cgo: refactor Go types a little bit

Code like this is not allowed by the upstream Go CGo implementation, but
was allowed by TinyGo:

    var _ int8 = C.int8_t(5)

The reason it shouldn't be allowed is a little bit complicated. While
it is true that C.int8_t is always the same underlying data type as Go
int8 (signed 8-bit integer), the C type is actually a typedef of one of
the base C types (usually unsigned char or signed char) which in turn do
_not_ map cleanly to Go types: the 'char' type is ambiguous (it may be
either signed or unsigned depending on the ABI) and types like 'int'
vary in size by ABI as well.

To make code more portable, I think it's better to match the upstream
implementation.

.circleci/config.yml

@@ -89,7 +89,7 @@ commands:
               # formatting of generated files.
               name: Check Go code formatting
               command: make fmt-check
-      - run: make gen-device
+      - run: make gen-device -j4
       - run: make smoketest XTENSA=0
       - save_cache:
           key: go-cache-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
@@ -98,12 +98,12 @@ commands:
             - /go/pkg/mod
 
 jobs:
-  test-llvm15-go118:
+  test-llvm14-go118:
     docker:
       - image: golang:1.18-buster
     steps:
       - test-linux:
-          llvm: "15"
+          llvm: "14"
     resource_class: large
 
 workflows:
@@ -111,4 +111,4 @@ workflows:
     jobs:
       # This tests our lowest supported versions of Go and LLVM, to make sure at
       # least the smoke tests still pass.
-      - test-llvm15-go118
+      - test-llvm14-go118

.github/workflows/build-macos.yml

@@ -88,7 +88,7 @@ jobs:
         if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
         run: make wasi-libc
       - name: make gen-device
-        run: make gen-device
+        run: make -j3 gen-device
       - name: Test TinyGo
         shell: bash
         run: make test GOTESTFLAGS="-short"

.github/workflows/linux.yml

@@ -255,7 +255,7 @@ jobs:
       - name: Build wasi-libc
         if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
         run: make wasi-libc
-      - run: make gen-device
+      - run: make gen-device -j4
       - name: Test TinyGo
         run: make ASSERT=1 test
       - name: Build TinyGo
@@ -267,7 +267,7 @@ jobs:
       - run: make smoketest
       - run: make wasmtest
       - run: make tinygo-baremetal
-  build-linux-cross:
+  build-linux-arm:
     # Build ARM Linux binaries, ready for release.
     # This intentionally uses an older Linux image, so that we compile against
     # an older glibc version and therefore are compatible with a wide range of
@@ -276,16 +276,6 @@ jobs:
     # in that process to avoid doing lots of duplicate work and to avoid
     # complications around precompiled libraries such as compiler-rt shipped as
     # part of the release tarball.
-    strategy:
-      matrix:
-        goarch: [ arm, arm64 ]
-        include:
-          - goarch: arm64
-            toolchain: aarch64-linux-gnu
-            libc: arm64
-          - goarch: arm
-            toolchain: arm-linux-gnueabihf
-            libc: armhf
     runs-on: ubuntu-20.04
     needs: build-linux
     steps:
@@ -296,8 +286,8 @@ jobs:
           sudo apt-get update
           sudo apt-get install --no-install-recommends \
               qemu-user \
-              g++-${{ matrix.toolchain }} \
+              g++-arm-linux-gnueabihf \
-              libc6-dev-${{ matrix.libc }}-cross
+              libc6-dev-armhf-cross
       - name: Install Go
         uses: actions/setup-go@v3
         with:
@@ -332,7 +322,7 @@ jobs:
         uses: actions/cache/restore@v3
         id: cache-llvm-build
         with:
-          key: llvm-build-16-linux-${{ matrix.goarch }}-v1
+          key: llvm-build-16-linux-arm-v1
           path: llvm-build
       - name: Build LLVM
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -343,7 +333,7 @@ jobs:
           # Install build dependencies.
           sudo apt-get install --no-install-recommends ninja-build
           # build!
-          make llvm-build CROSS=${{ matrix.toolchain }}
+          make llvm-build CROSS=arm-linux-gnueabihf
           # Remove unnecessary object files (to reduce cache size).
           find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
       - name: Save LLVM build cache
@@ -356,14 +346,14 @@ jobs:
         uses: actions/cache@v3
         id: cache-binaryen
         with:
-          key: binaryen-linux-${{ matrix.goarch }}-v1
+          key: binaryen-linux-arm-v1
           path: build/wasm-opt
       - name: Build Binaryen
         if: steps.cache-binaryen.outputs.cache-hit != 'true'
         run: |
           sudo apt-get install --no-install-recommends ninja-build
           git submodule update --init lib/binaryen
-          make CROSS=${{ matrix.toolchain }} binaryen
+          make CROSS=arm-linux-gnueabihf binaryen
       - name: Install fpm
         run: |
           sudo gem install --version 4.0.7 public_suffix
@@ -371,7 +361,7 @@ jobs:
           sudo gem install --no-document fpm
       - name: Build TinyGo binary
         run: |
-          make CROSS=${{ matrix.toolchain }}
+          make CROSS=arm-linux-gnueabihf
       - name: Download amd64 release
         uses: actions/download-artifact@v3
         with:
@@ -384,15 +374,129 @@ jobs:
         run: |
           cp -p build/tinygo   build/release/tinygo/bin
           cp -p build/wasm-opt build/release/tinygo/bin
-      - name: Create ${{ matrix.goarch }} release
+      - name: Create arm release
         run: |
-          make release deb RELEASEONLY=1 DEB_ARCH=${{ matrix.libc }}
+          make release deb RELEASEONLY=1 DEB_ARCH=armhf
-          cp -p build/release.tar.gz /tmp/tinygo.linux-${{ matrix.goarch }}.tar.gz
+          cp -p build/release.tar.gz /tmp/tinygo.linux-arm.tar.gz
-          cp -p build/release.deb    /tmp/tinygo_${{ matrix.libc }}.deb
+          cp -p build/release.deb    /tmp/tinygo_armhf.deb
       - name: Publish release artifact
         uses: actions/upload-artifact@v3
         with:
-          name: linux-${{ matrix.goarch }}-double-zipped
+          name: linux-arm-double-zipped
           path: |
-            /tmp/tinygo.linux-${{ matrix.goarch }}.tar.gz
+            /tmp/tinygo.linux-arm.tar.gz
-            /tmp/tinygo_${{ matrix.libc }}.deb
+            /tmp/tinygo_armhf.deb
+  build-linux-arm64:
+    # Build ARM64 Linux binaries, ready for release.
+    # It is set to "needs: build-linux" because it modifies the release created
+    # in that process to avoid doing lots of duplicate work and to avoid
+    # complications around precompiled libraries such as compiler-rt shipped as
+    # part of the release tarball.
+    runs-on: ubuntu-20.04
+    needs: build-linux
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Install apt dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install --no-install-recommends \
+              qemu-user \
+              g++-aarch64-linux-gnu \
+              libc6-dev-arm64-cross \
+              ninja-build
+      - name: Install Go
+        uses: actions/setup-go@v3
+        with:
+          go-version: '1.21'
+          cache: true
+      - name: Restore LLVM source cache
+        uses: actions/cache/restore@v3
+        id: cache-llvm-source
+        with:
+          key: llvm-source-16-linux-v1
+          path: |
+            llvm-project/clang/lib/Headers
+            llvm-project/clang/include
+            llvm-project/compiler-rt
+            llvm-project/lld/include
+            llvm-project/llvm/include
+      - name: Download LLVM source
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        run: make llvm-source
+      - name: Save LLVM source cache
+        uses: actions/cache/save@v3
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        with:
+          key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
+          path: |
+            llvm-project/clang/lib/Headers
+            llvm-project/clang/include
+            llvm-project/compiler-rt
+            llvm-project/lld/include
+            llvm-project/llvm/include
+      - name: Restore LLVM build cache
+        uses: actions/cache/restore@v3
+        id: cache-llvm-build
+        with:
+          key: llvm-build-16-linux-arm64-v1
+          path: llvm-build
+      - name: Build LLVM
+        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
+        run: |
+          # fetch LLVM source
+          rm -rf llvm-project
+          make llvm-source
+          # build!
+          make llvm-build CROSS=aarch64-linux-gnu
+          # Remove unnecessary object files (to reduce cache size).
+          find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
+      - name: Save LLVM build cache
+        uses: actions/cache/save@v3
+        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
+        with:
+          key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
+          path: llvm-build
+      - name: Cache Binaryen
+        uses: actions/cache@v3
+        id: cache-binaryen
+        with:
+          key: binaryen-linux-arm64-v1
+          path: build/wasm-opt
+      - name: Build Binaryen
+        if: steps.cache-binaryen.outputs.cache-hit != 'true'
+        run: |
+          git submodule update --init lib/binaryen
+          make CROSS=aarch64-linux-gnu binaryen
+      - name: Install fpm
+        run: |
+          sudo gem install --version 4.0.7 public_suffix
+          sudo gem install --version 2.7.6 dotenv
+          sudo gem install --no-document fpm
+      - name: Build TinyGo binary
+        run: |
+          make CROSS=aarch64-linux-gnu
+      - name: Download amd64 release
+        uses: actions/download-artifact@v3
+        with:
+          name: linux-amd64-double-zipped
+      - name: Extract amd64 release
+        run: |
+          mkdir -p build/release
+          tar -xf tinygo.linux-amd64.tar.gz -C build/release tinygo
+      - name: Modify release
+        run: |
+          cp -p build/tinygo   build/release/tinygo/bin
+          cp -p build/wasm-opt build/release/tinygo/bin
+      - name: Create arm64 release
+        run: |
+          make release deb RELEASEONLY=1 DEB_ARCH=arm64
+          cp -p build/release.tar.gz /tmp/tinygo.linux-arm64.tar.gz
+          cp -p build/release.deb    /tmp/tinygo_arm64.deb
+      - name: Publish release artifact
+        uses: actions/upload-artifact@v3
+        with:
+          name: linux-arm64-double-zipped
+          path: |
+            /tmp/tinygo.linux-arm64.tar.gz
+            /tmp/tinygo_arm64.deb

.github/workflows/sizediff.yml

@@ -49,7 +49,7 @@ jobs:
           path: |
             ~/.cache/go-build
             ~/go/pkg/mod
-      - run: make gen-device
+      - run: make gen-device -j4
       - name: Download drivers repo
         run: git clone https://github.com/tinygo-org/drivers.git
       - name: Save HEAD

.github/workflows/windows.yml

@@ -98,7 +98,7 @@ jobs:
         run: |
           scoop install wasmtime
       - name: make gen-device
-        run: make gen-device
+        run: make -j3 gen-device
       - name: Test TinyGo
         shell: bash
         run: make test GOTESTFLAGS="-short"

BUILDING.md

@@ -19,7 +19,6 @@ LLVM, Clang and LLD are quite light on dependencies, requiring only standard
 build tools to be built. Go is of course necessary to build TinyGo itself.
 
   * Go (1.18+)
-  * GNU Make
   * Standard build tools (gcc/clang)
   * git
   * CMake

Makefile

@@ -10,7 +10,7 @@ LLD_SRC ?= $(LLVM_PROJECTDIR)/lld
 
 # Try to autodetect LLVM build tools.
 # Versions are listed here in descending priority order.
-LLVM_VERSIONS = 16 15
+LLVM_VERSIONS = 16 15 14 13 12 11
 errifempty = $(if $(1),$(1),$(error $(2)))
 detect = $(shell which $(call errifempty,$(firstword $(foreach p,$(2),$(shell command -v $(p) 2> /dev/null && echo $(p)))),failed to locate $(1) at any of: $(2)))
 toolSearchPathsVersion = $(1)-$(2)
@@ -265,7 +265,7 @@ endif
 wasi-libc: lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a
 lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a:
 	@if [ ! -e lib/wasi-libc/Makefile ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
-	cd lib/wasi-libc && $(MAKE) -j4 EXTRA_CFLAGS="-O2 -g -DNDEBUG -mnontrapping-fptoint -msign-ext" MALLOC_IMPL=none CC=$(CLANG) AR=$(LLVM_AR) NM=$(LLVM_NM)
+	cd lib/wasi-libc && make -j4 EXTRA_CFLAGS="-O2 -g -DNDEBUG -mnontrapping-fptoint -msign-ext" MALLOC_IMPL=none CC=$(CLANG) AR=$(LLVM_AR) NM=$(LLVM_NM)
 
 # Check for Node.js used during WASM tests.
 NODEJS_VERSION := $(word 1,$(subst ., ,$(shell node -v | cut -c 2-)))
@@ -280,7 +280,7 @@ endif
 
 # Build the Go compiler.
 tinygo:
-	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  $(MAKE) llvm-source"; echo "  $(MAKE) $(LLVM_BUILDDIR)"; exit 1; fi
+	@if [ ! -f "$(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)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm osusergo" -ldflags="-X github.com/tinygo-org/tinygo/goenv.GitSha1=`git rev-parse --short HEAD`" .
 test: wasi-libc check-nodejs-version
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=20m -buildmode exe -tags "byollvm osusergo" ./builder ./cgo ./compileopts ./compiler ./interp ./transform .

builder/build.go

@@ -83,7 +83,8 @@ type packageAction struct {
 	FileHashes       map[string]string // hash of every file that's part of the package
 	EmbeddedFiles    map[string]string // hash of all the //go:embed files in the package
 	Imports          map[string]string // map from imported package to action ID hash
-	OptLevel         string            // LLVM optimization level (O0, O1, O2, Os, Oz)
+	OptLevel         int               // LLVM optimization level (0-3)
+	SizeLevel        int               // LLVM optimization for size level (0-2)
 	UndefinedGlobals []string          // globals that are left as external globals (no initializer)
 }
 
@@ -157,7 +158,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		return BuildResult{}, fmt.Errorf("unknown libc: %s", config.Target.Libc)
 	}
 
-	optLevel, speedLevel, sizeLevel := config.OptLevel()
+	optLevel, sizeLevel, _ := config.OptLevels()
 	compilerConfig := &compiler.Config{
 		Triple:          config.Triple(),
 		CPU:             config.CPU(),
@@ -320,6 +321,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					EmbeddedFiles:    make(map[string]string, len(allFiles)),
 					Imports:          make(map[string]string, len(pkg.Pkg.Imports())),
 					OptLevel:         optLevel,
+					SizeLevel:        sizeLevel,
 					UndefinedGlobals: undefinedGlobals,
 				}
 				for filePath, hash := range pkg.FileHashes {
@@ -530,13 +532,13 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			irbuilder := mod.Context().NewBuilder()
 			defer irbuilder.Dispose()
 			irbuilder.SetInsertPointAtEnd(block)
-			ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+			i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
 			for _, pkg := range lprogram.Sorted() {
 				pkgInit := mod.NamedFunction(pkg.Pkg.Path() + ".init")
 				if pkgInit.IsNil() {
 					panic("init not found for " + pkg.Pkg.Path())
 				}
-				irbuilder.CreateCall(pkgInit.GlobalValueType(), pkgInit, []llvm.Value{llvm.Undef(ptrType)}, "")
+				irbuilder.CreateCall(pkgInit.GlobalValueType(), pkgInit, []llvm.Value{llvm.Undef(i8ptrType)}, "")
 			}
 			irbuilder.CreateRetVoid()
 
@@ -741,17 +743,17 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			if config.GOOS() == "windows" {
 				// Options for the MinGW wrapper for the lld COFF linker.
 				ldflags = append(ldflags,
-					"-Xlink=/opt:lldlto="+strconv.Itoa(speedLevel),
+					"-Xlink=/opt:lldlto="+strconv.Itoa(optLevel),
 					"--thinlto-cache-dir="+filepath.Join(cacheDir, "thinlto"))
 			} else if config.GOOS() == "darwin" {
 				// Options for the ld64-compatible lld linker.
 				ldflags = append(ldflags,
-					"--lto-O"+strconv.Itoa(speedLevel),
+					"--lto-O"+strconv.Itoa(optLevel),
 					"-cache_path_lto", filepath.Join(cacheDir, "thinlto"))
 			} else {
 				// Options for the ELF linker.
 				ldflags = append(ldflags,
-					"--lto-O"+strconv.Itoa(speedLevel),
+					"--lto-O"+strconv.Itoa(optLevel),
 					"--thinlto-cache-dir="+filepath.Join(cacheDir, "thinlto"),
 				)
 			}
@@ -762,7 +764,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			if sizeLevel >= 2 {
 				// Workaround with roughly the same effect as
 				// https://reviews.llvm.org/D119342.
-				// Can hopefully be removed in LLVM 18.
+				// Can hopefully be removed in LLVM 15.
 				ldflags = append(ldflags,
 					"-mllvm", "--rotation-max-header-size=0")
 			}
@@ -1064,9 +1066,10 @@ func optimizeProgram(mod llvm.Module, config *compileopts.Config) error {
 		return err
 	}
 
-	// Run most of the whole-program optimizations (including the whole
+	// Optimization levels here are roughly the same as Clang, but probably not
-	// O0/O1/O2/Os/Oz optimization pipeline).
+	// exactly.
-	errs := transform.Optimize(mod, config)
+	optLevel, sizeLevel, inlinerThreshold := config.OptLevels()
+	errs := transform.Optimize(mod, config, optLevel, sizeLevel, inlinerThreshold)
 	if len(errs) > 0 {
 		return newMultiError(errs)
 	}

builder/library.go

@@ -178,6 +178,12 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 	if strings.HasPrefix(target, "riscv64-") {
 		args = append(args, "-march=rv64gc")
 	}
+	if strings.HasPrefix(target, "xtensa") {
+		// Hack to work around an issue in the Xtensa port:
+		// https://github.com/espressif/llvm-project/issues/52
+		// Hopefully this will be fixed soon (LLVM 14).
+		args = append(args, "-D__ELF__")
+	}
 
 	var once sync.Once
 

builder/musl.go

@@ -6,10 +6,12 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"strconv"
 	"strings"
 
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
+	"tinygo.org/x/go-llvm"
 )
 
 var Musl = Library{
@@ -91,7 +93,6 @@ var Musl = Library{
 			"-Wno-string-plus-int",
 			"-Wno-ignored-pragmas",
 			"-Wno-tautological-constant-out-of-range-compare",
-			"-Wno-deprecated-non-prototype",
 			"-Qunused-arguments",
 			// Select include dirs. Don't include standard library includes
 			// (that would introduce host dependencies and other complications),
@@ -105,6 +106,11 @@ var Musl = Library{
 			"-I" + muslDir + "/include",
 			"-fno-stack-protector",
 		}
+		llvmMajor, _ := strconv.Atoi(strings.SplitN(llvm.Version, ".", 2)[0])
+		if llvmMajor >= 15 {
+			// This flag was added in Clang 15. It is not present in LLVM 14.
+			cflags = append(cflags, "-Wno-deprecated-non-prototype")
+		}
 		return cflags
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/musl/src") },

builder/sizes_test.go

@@ -41,9 +41,9 @@ func TestBinarySize(t *testing.T) {
 	// This is a small number of very diverse targets that we want to test.
 	tests := []sizeTest{
 		// microcontrollers
-		{"hifive1b", "examples/echo", 4484, 280, 0, 2252},
+		{"hifive1b", "examples/echo", 4568, 280, 0, 2252},
-		{"microbit", "examples/serial", 2724, 388, 8, 2256},
+		{"microbit", "examples/serial", 2728, 388, 8, 2256},
-		{"wioterminal", "examples/pininterrupt", 6000, 1484, 116, 6816},
+		{"wioterminal", "examples/pininterrupt", 5996, 1484, 116, 6816},
 
 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the

cgo/cgo.go

@@ -75,20 +75,10 @@ type bitfieldInfo struct {
 }
 
 // cgoAliases list type aliases between Go and C, for types that are equivalent
-// in both languages. See addTypeAliases.
+// in both languages.
 var cgoAliases = map[string]string{
-	"C.int8_t":    "int8",
+	"float":  "float32",
-	"C.int16_t":   "int16",
+	"double": "float64",
-	"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",
-	"C.float":     "float32",
-	"C.double":    "float64",
-	"C._Bool":     "bool",
 }
 
 // builtinAliases are handled specially because they only exist on the Go side
@@ -314,9 +304,9 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 		cf := p.newCGoFile(f, i)
 		// These types are aliased with the corresponding types in C. For
 		// example, float in C is always float32 in Go.
-		cf.names["float"] = clangCursor{}
+		for name := range cgoAliases {
-		cf.names["double"] = clangCursor{}
+			cf.names[name] = clangCursor{}
-		cf.names["_Bool"] = clangCursor{}
+		}
 		// Now read all the names (identifies) that C defines in the header
 		// snippet.
 		cf.readNames(p.cgoHeaders[i], cflagsForCGo, filepath.Base(fset.File(f.Pos()).Name()), func(names map[string]clangCursor) {
@@ -1131,9 +1121,7 @@ func (p *cgoPackage) getUnnamedDeclName(prefix string, itf interface{}) string {
 // getASTDeclName will declare the given C AST node (if not already defined) and
 // will return its name, in the form of C.foo.
 func (f *cgoFile) getASTDeclName(name string, found clangCursor, iscall bool) string {
-	// Some types are defined in stdint.h and map directly to a particular Go
+	if alias := cgoAliases[name]; alias != "" {
-	// type.
-	if alias := cgoAliases["C."+name]; alias != "" {
 		return alias
 	}
 	node := f.getASTDeclNode(name, found, iscall)

cgo/libclang_config_llvm14.go

@@ -0,0 +1,15 @@
+//go:build !byollvm && llvm14
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/lib/llvm-14/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@14/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@14/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm14/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-14/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@14/lib -lclang -lffi
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@14/lib -lclang -lffi
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm14/lib -lclang
+*/
+import "C"

cgo/libclang_config_llvm16.go

@@ -1,4 +1,4 @@
-//go:build !byollvm && !llvm15
+//go:build !byollvm && !llvm14 && !llvm15
 
 package cgo
 

cgo/testdata/errors.go

@@ -22,8 +22,12 @@ import "C"
 // #warning another warning
 import "C"
 
+// #include <stdint.h>
+import "C"
+
 // Make sure that errors for the following lines won't change with future
 // additions to the CGo preamble.
+//
 //line errors.go:100
 var (
 	// constant too large
@@ -38,4 +42,8 @@ var (
 	_ byte = C.SOME_CONST_3
 
 	_ = C.SOME_CONST_4
+
+	// This must result in a type error. Previously, TinyGo would allow this
+	// code (which is not allowed by upstream Go).
+	_ int8 = C.int8_t(5)
 )

cgo/testdata/errors.out.go

@@ -11,6 +11,7 @@
 //     testdata/errors.go:108: undefined: C.SOME_CONST_1
 //     testdata/errors.go:110: cannot use C.SOME_CONST_3 (untyped int constant 1234) as byte value in variable declaration (overflows)
 //     testdata/errors.go:112: undefined: C.SOME_CONST_4
+//     testdata/errors.go:116: cannot use C.int8_t(5) (constant 5 of type C.schar) as int8 value in variable declaration
 
 package main
 
@@ -58,3 +59,5 @@ type C.struct_point_t struct {
 type C.point_t = C.struct_point_t
 
 const C.SOME_CONST_3 = 1234
+
+type C.int8_t = C.schar

compileopts/config.go

@@ -145,18 +145,18 @@ func (c *Config) Serial() string {
 
 // OptLevels returns the optimization level (0-2), size level (0-2), and inliner
 // threshold as used in the LLVM optimization pipeline.
-func (c *Config) OptLevel() (level string, speedLevel, sizeLevel int) {
+func (c *Config) OptLevels() (optLevel, sizeLevel int, inlinerThreshold uint) {
 	switch c.Options.Opt {
 	case "none", "0":
-		return "O0", 0, 0
+		return 0, 0, 0 // -O0
 	case "1":
-		return "O1", 1, 0
+		return 1, 0, 0 // -O1
 	case "2":
-		return "O2", 2, 0
+		return 2, 0, 225 // -O2
 	case "s":
-		return "Os", 2, 1
+		return 2, 1, 225 // -Os
 	case "z":
-		return "Oz", 2, 2 // default
+		return 2, 2, 5 // -Oz, default
 	default:
 		// This is not shown to the user: valid choices are already checked as
 		// part of Options.Verify(). It is here as a sanity check.

compiler/atomic.go

@@ -35,7 +35,17 @@ func (b *builder) createAtomicOp(name string) llvm.Value {
 	case "SwapInt32", "SwapInt64", "SwapUint32", "SwapUint64", "SwapUintptr", "SwapPointer":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
 		val := b.getValue(b.fn.Params[1], getPos(b.fn))
+		isPointer := val.Type().TypeKind() == llvm.PointerTypeKind
+		if isPointer {
+			// atomicrmw only supports integers, so cast to an integer.
+			// TODO: this is fixed in LLVM 15.
+			val = b.CreatePtrToInt(val, b.uintptrType, "")
+			ptr = b.CreateBitCast(ptr, llvm.PointerType(val.Type(), 0), "")
+		}
 		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpXchg, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
+		if isPointer {
+			oldVal = b.CreateIntToPtr(oldVal, b.i8ptrType, "")
+		}
 		return oldVal
 	case "CompareAndSwapInt32", "CompareAndSwapInt64", "CompareAndSwapUint32", "CompareAndSwapUint64", "CompareAndSwapUintptr", "CompareAndSwapPointer":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
@@ -53,6 +63,24 @@ func (b *builder) createAtomicOp(name string) llvm.Value {
 	case "StoreInt32", "StoreInt64", "StoreUint32", "StoreUint64", "StoreUintptr", "StorePointer":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
 		val := b.getValue(b.fn.Params[1], getPos(b.fn))
+		if strings.HasPrefix(b.Triple, "avr") {
+			// SelectionDAGBuilder is currently missing the "are unaligned atomics allowed" check for stores.
+			vType := val.Type()
+			isPointer := vType.TypeKind() == llvm.PointerTypeKind
+			if isPointer {
+				// libcalls only supports integers, so cast to an integer.
+				vType = b.uintptrType
+				val = b.CreatePtrToInt(val, vType, "")
+				ptr = b.CreateBitCast(ptr, llvm.PointerType(vType, 0), "")
+			}
+			name := fmt.Sprintf("__atomic_store_%d", vType.IntTypeWidth()/8)
+			fn := b.mod.NamedFunction(name)
+			if fn.IsNil() {
+				fn = llvm.AddFunction(b.mod, name, llvm.FunctionType(vType, []llvm.Type{ptr.Type(), vType, b.uintptrType}, false))
+			}
+			b.createCall(fn.GlobalValueType(), fn, []llvm.Value{ptr, val, llvm.ConstInt(b.uintptrType, 5, false)}, "")
+			return llvm.Value{}
+		}
 		store := b.CreateStore(val, ptr)
 		store.SetOrdering(llvm.AtomicOrderingSequentiallyConsistent)
 		store.SetAlignment(b.targetData.PrefTypeAlignment(val.Type())) // required

compiler/calls.go

@@ -45,7 +45,7 @@ func (b *builder) createRuntimeCallCommon(fnName string, args []llvm.Value, name
 	if llvmFn.IsNil() {
 		panic("trying to call non-existent function: " + fn.RelString(nil))
 	}
-	args = append(args, llvm.Undef(b.dataPtrType)) // unused context parameter
+	args = append(args, llvm.Undef(b.i8ptrType)) // unused context parameter
 	if isInvoke {
 		return b.createInvoke(fnType, llvmFn, args, name)
 	}

compiler/channel.go

@@ -33,27 +33,28 @@ func (b *builder) createChanSend(instr *ssa.Send) {
 	// store value-to-send
 	valueType := b.getLLVMType(instr.X.Type())
 	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
-	var valueAlloca, valueAllocaSize llvm.Value
+	var valueAlloca, valueAllocaCast, valueAllocaSize llvm.Value
 	if isZeroSize {
-		valueAlloca = llvm.ConstNull(b.dataPtrType)
+		valueAlloca = llvm.ConstNull(llvm.PointerType(valueType, 0))
+		valueAllocaCast = llvm.ConstNull(b.i8ptrType)
 	} else {
-		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
+		valueAlloca, valueAllocaCast, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
 		b.CreateStore(chanValue, valueAlloca)
 	}
 
 	// Allocate blockedlist buffer.
 	channelBlockedList := b.getLLVMRuntimeType("channelBlockedList")
-	channelBlockedListAlloca, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
+	channelBlockedListAlloca, channelBlockedListAllocaCast, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
 
 	// Do the send.
-	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAlloca, channelBlockedListAlloca}, "")
+	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
 
 	// End the lifetime of the allocas.
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
 	// https://bugs.llvm.org/show_bug.cgi?id=41742
-	b.emitLifetimeEnd(channelBlockedListAlloca, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
 	if !isZeroSize {
-		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
+		b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
 	}
 }
 
@@ -65,27 +66,28 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
 
 	// Allocate memory to receive into.
 	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
-	var valueAlloca, valueAllocaSize llvm.Value
+	var valueAlloca, valueAllocaCast, valueAllocaSize llvm.Value
 	if isZeroSize {
-		valueAlloca = llvm.ConstNull(b.dataPtrType)
+		valueAlloca = llvm.ConstNull(llvm.PointerType(valueType, 0))
+		valueAllocaCast = llvm.ConstNull(b.i8ptrType)
 	} else {
-		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
+		valueAlloca, valueAllocaCast, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
 	}
 
 	// Allocate blockedlist buffer.
 	channelBlockedList := b.getLLVMRuntimeType("channelBlockedList")
-	channelBlockedListAlloca, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
+	channelBlockedListAlloca, channelBlockedListAllocaCast, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
 
 	// Do the receive.
-	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAlloca, channelBlockedListAlloca}, "")
+	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
 	var received llvm.Value
 	if isZeroSize {
 		received = llvm.ConstNull(valueType)
 	} else {
 		received = b.CreateLoad(valueType, valueAlloca, "chan.received")
-		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
+		b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
 	}
-	b.emitLifetimeEnd(channelBlockedListAlloca, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
 
 	if unop.CommaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
@@ -157,7 +159,8 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 			sendValue := b.getValue(state.Send, state.Pos)
 			alloca := llvmutil.CreateEntryBlockAlloca(b.Builder, sendValue.Type(), "select.send.value")
 			b.CreateStore(sendValue, alloca)
-			selectState = b.CreateInsertValue(selectState, alloca, 1, "")
+			ptr := b.CreateBitCast(alloca, b.i8ptrType, "")
+			selectState = b.CreateInsertValue(selectState, ptr, 1, "")
 		default:
 			panic("unreachable")
 		}
@@ -165,10 +168,10 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	}
 
 	// Create a receive buffer, where the received value will be stored.
-	recvbuf := llvm.Undef(b.dataPtrType)
+	recvbuf := llvm.Undef(b.i8ptrType)
 	if recvbufSize != 0 {
 		allocaType := llvm.ArrayType(b.ctx.Int8Type(), int(recvbufSize))
-		recvbufAlloca, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
+		recvbufAlloca, _, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
 		recvbufAlloca.SetAlignment(recvbufAlign)
 		recvbuf = b.CreateGEP(allocaType, recvbufAlloca, []llvm.Value{
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
@@ -178,7 +181,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 
 	// Create the states slice (allocated on the stack).
 	statesAllocaType := llvm.ArrayType(chanSelectStateType, len(selectStates))
-	statesAlloca, statesSize := b.createTemporaryAlloca(statesAllocaType, "select.states.alloca")
+	statesAlloca, statesI8, statesSize := b.createTemporaryAlloca(statesAllocaType, "select.states.alloca")
 	for i, state := range selectStates {
 		// Set each slice element to the appropriate channel.
 		gep := b.CreateGEP(statesAllocaType, statesAlloca, []llvm.Value{
@@ -199,7 +202,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		// Stack-allocate operation structures.
 		// If these were simply created as a slice, they would heap-allocate.
 		chBlockAllocaType := llvm.ArrayType(b.getLLVMRuntimeType("channelBlockedList"), len(selectStates))
-		chBlockAlloca, chBlockSize := b.createTemporaryAlloca(chBlockAllocaType, "select.block.alloca")
+		chBlockAlloca, chBlockAllocaPtr, chBlockSize := b.createTemporaryAlloca(chBlockAllocaType, "select.block.alloca")
 		chBlockLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
 		chBlockPtr := b.CreateGEP(chBlockAllocaType, chBlockAlloca, []llvm.Value{
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
@@ -213,7 +216,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		}, "select.result")
 
 		// Terminate the lifetime of the operation structures.
-		b.emitLifetimeEnd(chBlockAlloca, chBlockSize)
+		b.emitLifetimeEnd(chBlockAllocaPtr, chBlockSize)
 	} else {
 		results = b.createRuntimeCall("tryChanSelect", []llvm.Value{
 			recvbuf,
@@ -222,7 +225,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	}
 
 	// Terminate the lifetime of the states alloca.
-	b.emitLifetimeEnd(statesAlloca, statesSize)
+	b.emitLifetimeEnd(statesI8, statesSize)
 
 	// The result value does not include all the possible received values,
 	// because we can't load them in advance. Instead, the *ssa.Extract
@@ -262,6 +265,7 @@ func (b *builder) getChanSelectResult(expr *ssa.Extract) llvm.Value {
 		// it to the correct type, and dereference it.
 		recvbuf := b.selectRecvBuf[expr.Tuple.(*ssa.Select)]
 		typ := b.getLLVMType(expr.Type())
-		return b.CreateLoad(typ, recvbuf, "")
+		ptr := b.CreateBitCast(recvbuf, llvm.PointerType(typ, 0), "")
+		return b.CreateLoad(typ, ptr, "")
 	}
 }

compiler/compiler.go

@@ -75,9 +75,10 @@ type compilerContext struct {
 	machine          llvm.TargetMachine
 	targetData       llvm.TargetData
 	intType          llvm.Type
-	dataPtrType      llvm.Type // pointer in address space 0
+	i8ptrType        llvm.Type // for convenience
-	funcPtrType      llvm.Type // pointer in function address space (1 for AVR, 0 elsewhere)
+	rawVoidFuncType  llvm.Type // for convenience
 	funcPtrAddrSpace int
+	hasTypedPointers bool // for LLVM 14 backwards compatibility
 	uintptrType      llvm.Type
 	program          *ssa.Program
 	diagnostics      []error
@@ -122,12 +123,13 @@ func newCompilerContext(moduleName string, machine llvm.TargetMachine, config *C
 	} else {
 		panic("unknown pointer size")
 	}
-	c.dataPtrType = llvm.PointerType(c.ctx.Int8Type(), 0)
+	c.i8ptrType = llvm.PointerType(c.ctx.Int8Type(), 0)
 
 	dummyFuncType := llvm.FunctionType(c.ctx.VoidType(), nil, false)
 	dummyFunc := llvm.AddFunction(c.mod, "tinygo.dummy", dummyFuncType)
 	c.funcPtrAddrSpace = dummyFunc.Type().PointerAddressSpace()
-	c.funcPtrType = dummyFunc.Type()
+	c.hasTypedPointers = c.i8ptrType != llvm.PointerType(c.ctx.Int16Type(), 0) // with opaque pointers, all pointers are the same type (LLVM 15+)
+	c.rawVoidFuncType = dummyFunc.Type()
 	dummyFunc.EraseFromParentAsFunction()
 
 	return c
@@ -415,14 +417,16 @@ func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 		case types.Uintptr:
 			return c.uintptrType
 		case types.UnsafePointer:
-			return c.dataPtrType
+			return c.i8ptrType
 		default:
 			panic("unknown basic type: " + typ.String())
 		}
-	case *types.Chan, *types.Map, *types.Pointer:
+	case *types.Chan:
-		return c.dataPtrType // all pointers are the same
+		return llvm.PointerType(c.getLLVMRuntimeType("channel"), 0)
 	case *types.Interface:
 		return c.getLLVMRuntimeType("_interface")
+	case *types.Map:
+		return llvm.PointerType(c.getLLVMRuntimeType("hashmap"), 0)
 	case *types.Named:
 		if st, ok := typ.Underlying().(*types.Struct); ok {
 			// Structs are a special case. While other named types are ignored
@@ -437,11 +441,21 @@ func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 			return llvmType
 		}
 		return c.getLLVMType(typ.Underlying())
+	case *types.Pointer:
+		if c.hasTypedPointers {
+			ptrTo := c.getLLVMType(typ.Elem())
+			return llvm.PointerType(ptrTo, 0)
+		}
+		return c.i8ptrType // all pointers are the same
 	case *types.Signature: // function value
 		return c.getFuncType(typ)
 	case *types.Slice:
+		ptrType := c.i8ptrType
+		if c.hasTypedPointers {
+			ptrType = llvm.PointerType(c.getLLVMType(typ.Elem()), 0)
+		}
 		members := []llvm.Type{
-			c.dataPtrType,
+			ptrType,
 			c.uintptrType, // len
 			c.uintptrType, // cap
 		}
@@ -531,8 +545,8 @@ func (c *compilerContext) createDIType(typ types.Type) llvm.Metadata {
 				Elements: []llvm.Metadata{
 					c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
 						Name:         "ptr",
-						SizeInBits:   c.targetData.TypeAllocSize(c.dataPtrType) * 8,
+						SizeInBits:   c.targetData.TypeAllocSize(c.i8ptrType) * 8,
-						AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.dataPtrType)) * 8,
+						AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.i8ptrType)) * 8,
 						OffsetInBits: 0,
 						Type:         c.getDIType(types.NewPointer(types.Typ[types.Byte])),
 					}),
@@ -1534,18 +1548,21 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		src := argValues[0]
 		elems := argValues[1]
 		srcBuf := b.CreateExtractValue(src, 0, "append.srcBuf")
+		srcPtr := b.CreateBitCast(srcBuf, b.i8ptrType, "append.srcPtr")
 		srcLen := b.CreateExtractValue(src, 1, "append.srcLen")
 		srcCap := b.CreateExtractValue(src, 2, "append.srcCap")
 		elemsBuf := b.CreateExtractValue(elems, 0, "append.elemsBuf")
+		elemsPtr := b.CreateBitCast(elemsBuf, b.i8ptrType, "append.srcPtr")
 		elemsLen := b.CreateExtractValue(elems, 1, "append.elemsLen")
 		elemType := b.getLLVMType(argTypes[0].Underlying().(*types.Slice).Elem())
 		elemSize := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(elemType), false)
-		result := b.createRuntimeCall("sliceAppend", []llvm.Value{srcBuf, elemsBuf, srcLen, srcCap, elemsLen, elemSize}, "append.new")
+		result := b.createRuntimeCall("sliceAppend", []llvm.Value{srcPtr, elemsPtr, srcLen, srcCap, elemsLen, elemSize}, "append.new")
 		newPtr := b.CreateExtractValue(result, 0, "append.newPtr")
+		newBuf := b.CreateBitCast(newPtr, srcBuf.Type(), "append.newBuf")
 		newLen := b.CreateExtractValue(result, 1, "append.newLen")
 		newCap := b.CreateExtractValue(result, 2, "append.newCap")
 		newSlice := llvm.Undef(src.Type())
-		newSlice = b.CreateInsertValue(newSlice, newPtr, 0, "")
+		newSlice = b.CreateInsertValue(newSlice, newBuf, 0, "")
 		newSlice = b.CreateInsertValue(newSlice, newLen, 1, "")
 		newSlice = b.CreateInsertValue(newSlice, newCap, 2, "")
 		return newSlice, nil
@@ -1593,6 +1610,9 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 
 			// The pointer to the data to be cleared.
 			llvmBuf := b.CreateExtractValue(value, 0, "buf")
+			if llvmBuf.Type() != b.i8ptrType { // compatibility with LLVM 14
+				llvmBuf = b.CreateBitCast(llvmBuf, b.i8ptrType, "")
+			}
 
 			// The length (in bytes) to be cleared.
 			llvmLen := b.CreateExtractValue(value, 1, "len")
@@ -1627,6 +1647,8 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		dstBuf := b.CreateExtractValue(dst, 0, "copy.dstArray")
 		srcBuf := b.CreateExtractValue(src, 0, "copy.srcArray")
 		elemType := b.getLLVMType(argTypes[0].Underlying().(*types.Slice).Elem())
+		dstBuf = b.CreateBitCast(dstBuf, b.i8ptrType, "copy.dstPtr")
+		srcBuf = b.CreateBitCast(srcBuf, b.i8ptrType, "copy.srcPtr")
 		elemSize := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(elemType), false)
 		return b.createRuntimeCall("sliceCopy", []llvm.Value{dstBuf, srcBuf, dstLen, srcLen, elemSize}, "copy.n"), nil
 	case "delete":
@@ -1866,13 +1888,14 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		switch value := instr.Value.(type) {
 		case *ssa.Function:
 			// Regular function call. No context is necessary.
-			context = llvm.Undef(b.dataPtrType)
+			context = llvm.Undef(b.i8ptrType)
 			if info.variadic && len(fn.Params) == 0 {
 				// This matches Clang, see: https://godbolt.org/z/Gqv49xKMq
 				// Eventually we might be able to eliminate this special case
 				// entirely. For details, see:
 				// https://discourse.llvm.org/t/rfc-enabling-wstrict-prototypes-by-default-in-c/60521
 				calleeType = llvm.FunctionType(callee.GlobalValueType().ReturnType(), nil, false)
+				callee = llvm.ConstBitCast(callee, llvm.PointerType(calleeType, b.funcPtrAddrSpace))
 			}
 		case *ssa.MakeClosure:
 			// A call on a func value, but the callee is trivial to find. For
@@ -1900,14 +1923,13 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		params = append(params, typecode)
 		callee = b.getInvokeFunction(instr)
 		calleeType = callee.GlobalValueType()
-		context = llvm.Undef(b.dataPtrType)
+		context = llvm.Undef(b.i8ptrType)
 	} else {
 		// Function pointer.
 		value := b.getValue(instr.Value, getPos(instr))
 		// This is a func value, which cannot be called directly. We have to
 		// extract the function pointer and context first from the func value.
-		callee, context = b.decodeFuncValue(value)
+		calleeType, callee, context = b.decodeFuncValue(value, instr.Value.Type().Underlying().(*types.Signature))
-		calleeType = b.getLLVMFunctionType(instr.Value.Type().Underlying().(*types.Signature))
 		b.createNilCheck(instr.Value, callee, "fpcall")
 	}
 
@@ -1940,7 +1962,7 @@ func (b *builder) getValue(expr ssa.Value, pos token.Pos) llvm.Value {
 			return llvm.Undef(b.getLLVMType(expr.Type()))
 		}
 		_, fn := b.getFunction(expr)
-		return b.createFuncValue(fn, llvm.Undef(b.dataPtrType), expr.Signature)
+		return b.createFuncValue(fn, llvm.Undef(b.i8ptrType), expr.Signature)
 	case *ssa.Global:
 		value := b.getGlobal(expr)
 		if value.IsNil() {
@@ -2008,6 +2030,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			sizeValue := llvm.ConstInt(b.uintptrType, size, false)
 			layoutValue := b.createObjectLayout(typ, expr.Pos())
 			buf := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, layoutValue}, expr.Comment)
+			buf = b.CreateBitCast(buf, llvm.PointerType(typ, 0), "")
 			return buf, nil
 		} else {
 			buf := llvmutil.CreateEntryBlockAlloca(b.Builder, typ, expr.Comment)
@@ -2053,6 +2076,10 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 				value = b.CreateInsertValue(value, field, i, "changetype.struct")
 			}
 			return value, nil
+		case llvm.PointerTypeKind:
+			// This can happen with pointers to structs. This case is easy:
+			// simply bitcast the pointer to the destination type.
+			return b.CreateBitCast(x, llvmType, "changetype.pointer"), nil
 		default:
 			return llvm.Value{}, errors.New("todo: unknown ChangeType type: " + expr.X.Type().String())
 		}
@@ -2129,12 +2156,12 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			// Can't load directly from array (as index is non-constant), so
 			// have to do it using an alloca+gep+load.
 			arrayType := collection.Type()
-			alloca, allocaSize := b.createTemporaryAlloca(arrayType, "index.alloca")
+			alloca, allocaPtr, allocaSize := b.createTemporaryAlloca(arrayType, "index.alloca")
 			b.CreateStore(collection, alloca)
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			ptr := b.CreateInBoundsGEP(arrayType, alloca, []llvm.Value{zero, index}, "index.gep")
 			result := b.CreateLoad(arrayType.ElementType(), ptr, "index.load")
-			b.emitLifetimeEnd(alloca, allocaSize)
+			b.emitLifetimeEnd(allocaPtr, allocaSize)
 			return result, nil
 		default:
 			panic("unknown *ssa.Index type")
@@ -2237,6 +2264,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		sliceSize := b.CreateBinOp(llvm.Mul, elemSizeValue, sliceCapCast, "makeslice.cap")
 		layoutValue := b.createObjectLayout(llvmElemType, expr.Pos())
 		slicePtr := b.createRuntimeCall("alloc", []llvm.Value{sliceSize, layoutValue}, "makeslice.buf")
+		slicePtr = b.CreateBitCast(slicePtr, llvm.PointerType(llvmElemType, 0), "makeslice.array")
 
 		// Extend or truncate if necessary. This is safe as we've already done
 		// the bounds check.
@@ -2282,7 +2310,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		default:
 			panic("unknown type in range: " + typ.String())
 		}
-		it, _ := b.createTemporaryAlloca(iteratorType, "range.it")
+		it, _, _ := b.createTemporaryAlloca(iteratorType, "range.it")
 		b.CreateStore(llvm.ConstNull(iteratorType), it)
 		return it, nil
 	case *ssa.Select:
@@ -2450,6 +2478,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		arrayLen := expr.Type().Underlying().(*types.Pointer).Elem().Underlying().(*types.Array).Len()
 		b.createSliceToArrayPointerCheck(sliceLen, arrayLen)
 		ptr := b.CreateExtractValue(slice, 0, "")
+		ptr = b.CreateBitCast(ptr, b.getLLVMType(expr.Type()), "")
 		return ptr, nil
 	case *ssa.TypeAssert:
 		return b.createTypeAssert(expr), nil
@@ -2915,16 +2944,16 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 				zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 				strPtr = llvm.ConstInBoundsGEP(globalType, global, []llvm.Value{zero, zero})
 			} else {
-				strPtr = llvm.ConstNull(c.dataPtrType)
+				strPtr = llvm.ConstNull(c.i8ptrType)
 			}
 			strObj := llvm.ConstNamedStruct(c.getLLVMRuntimeType("_string"), []llvm.Value{strPtr, strLen})
 			return strObj
 		} else if typ.Kind() == types.UnsafePointer {
 			if !expr.IsNil() {
 				value, _ := constant.Uint64Val(constant.ToInt(expr.Value))
-				return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, value, false), c.dataPtrType)
+				return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, value, false), c.i8ptrType)
 			}
-			return llvm.ConstNull(c.dataPtrType)
+			return llvm.ConstNull(c.i8ptrType)
 		} else if typ.Info()&types.IsUnsigned != 0 {
 			n, _ := constant.Uint64Val(constant.ToInt(expr.Value))
 			return llvm.ConstInt(llvmType, n, false)
@@ -2968,7 +2997,7 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 		// Create a generic nil interface with no dynamic type (typecode=0).
 		fields := []llvm.Value{
 			llvm.ConstInt(c.uintptrType, 0, false),
-			llvm.ConstPointerNull(c.dataPtrType),
+			llvm.ConstPointerNull(c.i8ptrType),
 		}
 		return llvm.ConstNamedStruct(c.getLLVMRuntimeType("_interface"), fields)
 	case *types.Pointer:
@@ -2985,7 +3014,8 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 		if expr.Value != nil {
 			panic("expected nil slice constant")
 		}
-		llvmPtr := llvm.ConstPointerNull(c.dataPtrType)
+		elemType := c.getLLVMType(typ.Elem())
+		llvmPtr := llvm.ConstPointerNull(llvm.PointerType(elemType, 0))
 		llvmLen := llvm.ConstInt(c.uintptrType, 0, false)
 		slice := c.ctx.ConstStruct([]llvm.Value{
 			llvmPtr, // backing array
@@ -3026,7 +3056,7 @@ func (b *builder) createConvert(typeFrom, typeTo types.Type, value llvm.Value, p
 
 	// Conversion between pointers and unsafe.Pointer.
 	if isPtrFrom && isPtrTo {
-		return value, nil
+		return b.CreateBitCast(value, llvmTypeTo, ""), nil
 	}
 
 	switch typeTo := typeTo.Underlying().(type) {
@@ -3265,7 +3295,7 @@ func (b *builder) createUnOp(unop *ssa.UnOp) (llvm.Value, error) {
 			if fn.IsNil() {
 				return llvm.Value{}, b.makeError(unop.Pos(), "cgo function not found: "+name)
 			}
-			return fn, nil
+			return b.CreateBitCast(fn, b.i8ptrType, ""), nil
 		} else {
 			b.createNilCheck(unop.X, x, "deref")
 			load := b.CreateLoad(valueType, x, "")

compiler/compiler_test.go

@@ -91,12 +91,14 @@ func TestCompiler(t *testing.T) {
 			}
 
 			// Optimize IR a little.
-			passOptions := llvm.NewPassBuilderOptions()
+			funcPasses := llvm.NewFunctionPassManagerForModule(mod)
-			defer passOptions.Dispose()
+			defer funcPasses.Dispose()
-			err = mod.RunPasses("instcombine", llvm.TargetMachine{}, passOptions)
+			funcPasses.AddInstructionCombiningPass()
-			if err != nil {
+			funcPasses.InitializeFunc()
-				t.Error(err)
+			for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
+				funcPasses.RunFunc(fn)
 			}
+			funcPasses.FinalizeFunc()
 
 			outFilePrefix := tc.file[:len(tc.file)-3]
 			if tc.target != "" {
@@ -167,9 +169,9 @@ func filterIrrelevantIRLines(lines []string) []string {
 		if strings.HasPrefix(line, "source_filename = ") {
 			continue
 		}
-		if llvmVersion < 15 && strings.HasPrefix(line, "target datalayout = ") {
+		if llvmVersion < 14 && strings.HasPrefix(line, "target datalayout = ") {
 			// The datalayout string may vary betewen LLVM versions.
-			// Right now test outputs are for LLVM 15 and higher.
+			// Right now test outputs are for LLVM 14 and higher.
 			continue
 		}
 		out = append(out, line)

compiler/defer.go

@@ -60,8 +60,9 @@ func (b *builder) deferInitFunc() {
 	b.deferBuiltinFuncs = make(map[ssa.Value]deferBuiltin)
 
 	// Create defer list pointer.
-	b.deferPtr = b.CreateAlloca(b.dataPtrType, "deferPtr")
+	deferType := llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)
-	b.CreateStore(llvm.ConstPointerNull(b.dataPtrType), b.deferPtr)
+	b.deferPtr = b.CreateAlloca(deferType, "deferPtr")
+	b.CreateStore(llvm.ConstPointerNull(deferType), b.deferPtr)
 
 	if b.hasDeferFrame() {
 		// Set up the defer frame with the current stack pointer.
@@ -248,7 +249,8 @@ func isInLoop(start *ssa.BasicBlock) bool {
 func (b *builder) createDefer(instr *ssa.Defer) {
 	// The pointer to the previous defer struct, which we will replace to
 	// make a linked list.
-	next := b.CreateLoad(b.dataPtrType, b.deferPtr, "defer.next")
+	deferType := llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)
+	next := b.CreateLoad(deferType, b.deferPtr, "defer.next")
 
 	var values []llvm.Value
 	valueTypes := []llvm.Type{b.uintptrType, next.Type()}
@@ -269,7 +271,7 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		typecode := b.CreateExtractValue(itf, 0, "invoke.func.typecode")
 		receiverValue := b.CreateExtractValue(itf, 1, "invoke.func.receiver")
 		values = []llvm.Value{callback, next, typecode, receiverValue}
-		valueTypes = append(valueTypes, b.dataPtrType, b.dataPtrType)
+		valueTypes = append(valueTypes, b.i8ptrType, b.i8ptrType)
 		for _, arg := range instr.Call.Args {
 			val := b.getValue(arg, getPos(instr))
 			values = append(values, val)
@@ -389,8 +391,9 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		// This may be hit a variable number of times, so use a heap allocation.
 		size := b.targetData.TypeAllocSize(deferredCallType)
 		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
-		nilPtr := llvm.ConstNull(b.dataPtrType)
+		nilPtr := llvm.ConstNull(b.i8ptrType)
-		alloca = b.createRuntimeCall("alloc", []llvm.Value{sizeValue, nilPtr}, "defer.alloc.call")
+		allocCall := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, nilPtr}, "defer.alloc.call")
+		alloca = b.CreateBitCast(allocCall, llvm.PointerType(deferredCallType, 0), "defer.alloc")
 	}
 	if b.NeedsStackObjects {
 		b.trackPointer(alloca)
@@ -398,12 +401,14 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 	b.CreateStore(deferredCall, alloca)
 
 	// Push it on top of the linked list by replacing deferPtr.
-	b.CreateStore(alloca, b.deferPtr)
+	allocaCast := b.CreateBitCast(alloca, next.Type(), "defer.alloca.cast")
+	b.CreateStore(allocaCast, b.deferPtr)
 }
 
 // createRunDefers emits code to run all deferred functions.
 func (b *builder) createRunDefers() {
 	deferType := b.getLLVMRuntimeType("_defer")
+	deferPtrType := llvm.PointerType(deferType, 0)
 
 	// Add a loop like the following:
 	//     for stack != nil {
@@ -430,7 +435,7 @@ func (b *builder) createRunDefers() {
 	// Create loop head:
 	//     for stack != nil {
 	b.SetInsertPointAtEnd(loophead)
-	deferData := b.CreateLoad(b.dataPtrType, b.deferPtr, "")
+	deferData := b.CreateLoad(deferPtrType, b.deferPtr, "")
 	stackIsNil := b.CreateICmp(llvm.IntEQ, deferData, llvm.ConstPointerNull(deferData.Type()), "stackIsNil")
 	b.CreateCondBr(stackIsNil, end, loop)
 
@@ -443,7 +448,7 @@ func (b *builder) createRunDefers() {
 		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 		llvm.ConstInt(b.ctx.Int32Type(), 1, false), // .next field
 	}, "stack.next.gep")
-	nextStack := b.CreateLoad(b.dataPtrType, nextStackGEP, "stack.next")
+	nextStack := b.CreateLoad(deferPtrType, nextStackGEP, "stack.next")
 	b.CreateStore(nextStack, b.deferPtr)
 	gep := b.CreateInBoundsGEP(deferType, deferData, []llvm.Value{
 		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
@@ -464,26 +469,28 @@ func (b *builder) createRunDefers() {
 			// Call on an value or interface value.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
 
 			if !callback.IsInvoke() {
 				//Expect funcValue to be passed through the deferred call.
 				valueTypes = append(valueTypes, b.getFuncType(callback.Signature()))
 			} else {
 				//Expect typecode
-				valueTypes = append(valueTypes, b.dataPtrType, b.dataPtrType)
+				valueTypes = append(valueTypes, b.i8ptrType, b.i8ptrType)
 			}
 
 			for _, arg := range callback.Args {
 				valueTypes = append(valueTypes, b.getLLVMType(arg.Type()))
 			}
 
+			deferredCallType := b.ctx.StructType(valueTypes, false)
+			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
+
 			// Extract the params from the struct (including receiver).
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
-			deferredCallType := b.ctx.StructType(valueTypes, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "gep")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "gep")
 				forwardParam := b.CreateLoad(valueTypes[i], gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -498,8 +505,7 @@ func (b *builder) createRunDefers() {
 
 				//Get function pointer and context
 				var context llvm.Value
-				fnPtr, context = b.decodeFuncValue(funcValue)
+				fnType, fnPtr, context = b.decodeFuncValue(funcValue, callback.Signature())
-				fnType = b.getLLVMFunctionType(callback.Signature())
 
 				//Pass context
 				forwardParams = append(forwardParams, context)
@@ -513,7 +519,7 @@ func (b *builder) createRunDefers() {
 				// Add the context parameter. An interface call cannot also be a
 				// closure but we have to supply the parameter anyway for platforms
 				// with a strict calling convention.
-				forwardParams = append(forwardParams, llvm.Undef(b.dataPtrType))
+				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
 			}
 
 			b.createCall(fnType, fnPtr, forwardParams, "")
@@ -522,17 +528,18 @@ func (b *builder) createRunDefers() {
 			// Direct call.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
 			for _, param := range getParams(callback.Signature) {
 				valueTypes = append(valueTypes, b.getLLVMType(param.Type()))
 			}
 			deferredCallType := b.ctx.StructType(valueTypes, false)
+			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
 
 			// Extract the params from the struct.
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := range getParams(callback.Signature) {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
 				forwardParam := b.CreateLoad(valueTypes[i+2], gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -542,7 +549,7 @@ func (b *builder) createRunDefers() {
 			if !b.getFunctionInfo(callback).exported {
 				// Add the context parameter. We know it is ignored by the receiving
 				// function, but we have to pass one anyway.
-				forwardParams = append(forwardParams, llvm.Undef(b.dataPtrType))
+				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
 			}
 
 			// Call real function.
@@ -552,19 +559,20 @@ func (b *builder) createRunDefers() {
 		case *ssa.MakeClosure:
 			// Get the real defer struct type and cast to it.
 			fn := callback.Fn.(*ssa.Function)
-			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
 			params := fn.Signature.Params()
 			for i := 0; i < params.Len(); i++ {
 				valueTypes = append(valueTypes, b.getLLVMType(params.At(i).Type()))
 			}
-			valueTypes = append(valueTypes, b.dataPtrType) // closure
+			valueTypes = append(valueTypes, b.i8ptrType) // closure
 			deferredCallType := b.ctx.StructType(valueTypes, false)
+			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
 
 			// Extract the params from the struct.
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "")
 				forwardParam := b.CreateLoad(valueTypes[i], gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -576,7 +584,7 @@ func (b *builder) createRunDefers() {
 			db := b.deferBuiltinFuncs[callback]
 
 			//Get parameter types
-			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
 
 			//Get signature from call results
 			params := callback.Type().Underlying().(*types.Signature).Params()
@@ -585,12 +593,13 @@ func (b *builder) createRunDefers() {
 			}
 
 			deferredCallType := b.ctx.StructType(valueTypes, false)
+			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
 
 			// Extract the params from the struct.
 			var argValues []llvm.Value
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := 0; i < params.Len(); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
 				forwardParam := b.CreateLoad(valueTypes[i+2], gep, "param")
 				argValues = append(argValues, forwardParam)
 			}

compiler/func.go

@@ -13,14 +13,35 @@ import (
 // createFuncValue creates a function value from a raw function pointer with no
 // context.
 func (b *builder) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
+	return b.compilerContext.createFuncValue(b.Builder, funcPtr, context, sig)
+}
+
+// createFuncValue creates a function value from a raw function pointer with no
+// context.
+func (c *compilerContext) createFuncValue(builder llvm.Builder, funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
 	// Closure is: {context, function pointer}
-	funcValueType := b.getFuncType(sig)
+	funcValueScalar := llvm.ConstBitCast(funcPtr, c.rawVoidFuncType)
+	funcValueType := c.getFuncType(sig)
 	funcValue := llvm.Undef(funcValueType)
-	funcValue = b.CreateInsertValue(funcValue, context, 0, "")
+	funcValue = builder.CreateInsertValue(funcValue, context, 0, "")
-	funcValue = b.CreateInsertValue(funcValue, funcPtr, 1, "")
+	funcValue = builder.CreateInsertValue(funcValue, funcValueScalar, 1, "")
 	return funcValue
 }
 
+// getFuncSignatureID returns a new external global for a given signature. This
+// global reference is not real, it is only used during func lowering to assign
+// signature types to functions and will then be removed.
+func (c *compilerContext) getFuncSignatureID(sig *types.Signature) llvm.Value {
+	s, _ := getTypeCodeName(sig)
+	sigGlobalName := "reflect/types.funcid:" + s
+	sigGlobal := c.mod.NamedGlobal(sigGlobalName)
+	if sigGlobal.IsNil() {
+		sigGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), sigGlobalName)
+		sigGlobal.SetGlobalConstant(true)
+	}
+	return sigGlobal
+}
+
 // extractFuncScalar returns some scalar that can be used in comparisons. It is
 // a cheap operation.
 func (b *builder) extractFuncScalar(funcValue llvm.Value) llvm.Value {
@@ -34,20 +55,28 @@ func (b *builder) extractFuncContext(funcValue llvm.Value) llvm.Value {
 }
 
 // decodeFuncValue extracts the context and the function pointer from this func
-// value.
+// value. This may be an expensive operation.
-func (b *builder) decodeFuncValue(funcValue llvm.Value) (funcPtr, context llvm.Value) {
+func (b *builder) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (funcType llvm.Type, funcPtr, context llvm.Value) {
 	context = b.CreateExtractValue(funcValue, 0, "")
 	funcPtr = b.CreateExtractValue(funcValue, 1, "")
+	if !funcPtr.IsAConstantExpr().IsNil() && funcPtr.Opcode() == llvm.BitCast {
+		funcPtr = funcPtr.Operand(0) // needed for LLVM 14 (no opaque pointers)
+	}
+	if sig != nil {
+		funcType = b.getRawFuncType(sig)
+		llvmSig := llvm.PointerType(funcType, b.funcPtrAddrSpace)
+		funcPtr = b.CreateBitCast(funcPtr, llvmSig, "")
+	}
 	return
 }
 
 // getFuncType returns the type of a func value given a signature.
 func (c *compilerContext) getFuncType(typ *types.Signature) llvm.Type {
-	return c.ctx.StructType([]llvm.Type{c.dataPtrType, c.funcPtrType}, false)
+	return c.ctx.StructType([]llvm.Type{c.i8ptrType, c.rawVoidFuncType}, false)
 }
 
-// getLLVMFunctionType returns a LLVM function type for a given signature.
+// getRawFuncType returns a LLVM function type for a given signature.
-func (c *compilerContext) getLLVMFunctionType(typ *types.Signature) llvm.Type {
+func (c *compilerContext) getRawFuncType(typ *types.Signature) llvm.Type {
 	// Get the return type.
 	var returnType llvm.Type
 	switch typ.Results().Len() {
@@ -75,7 +104,7 @@ func (c *compilerContext) getLLVMFunctionType(typ *types.Signature) llvm.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.
-			recv = c.dataPtrType
+			recv = c.i8ptrType
 		}
 		for _, info := range c.expandFormalParamType(recv, "", nil) {
 			paramTypes = append(paramTypes, info.llvmType)
@@ -88,7 +117,7 @@ func (c *compilerContext) getLLVMFunctionType(typ *types.Signature) llvm.Type {
 		}
 	}
 	// All functions take these parameters at the end.
-	paramTypes = append(paramTypes, c.dataPtrType) // context
+	paramTypes = append(paramTypes, c.i8ptrType) // context
 
 	// Make a func type out of the signature.
 	return llvm.FunctionType(returnType, paramTypes, false)

compiler/gc.go

@@ -81,6 +81,9 @@ func (b *builder) trackValue(value llvm.Value) {
 // trackPointer creates a call to runtime.trackPointer, bitcasting the poitner
 // first if needed. The input value must be of LLVM pointer type.
 func (b *builder) trackPointer(value llvm.Value) {
+	if value.Type() != b.i8ptrType {
+		value = b.CreateBitCast(value, b.i8ptrType, "")
+	}
 	b.createRuntimeCall("trackPointer", []llvm.Value{value, b.stackChainAlloca}, "")
 }
 

compiler/goroutine.go

@@ -21,7 +21,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 
 	var prefix string
 	var funcPtr llvm.Value
-	var funcType llvm.Type
+	var funcPtrType llvm.Type
 	hasContext := false
 	if callee := instr.Call.StaticCallee(); callee != nil {
 		// Static callee is known. This makes it easier to start a new
@@ -42,7 +42,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 			params = append(params, context) // context parameter
 			hasContext = true
 		}
-		funcType, funcPtr = b.getFunction(callee)
+		funcPtrType, funcPtr = b.getFunction(callee)
 	} else if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
 		// We cheat. None of the builtins do any long or blocking operation, so
 		// we might as well run these builtins right away without the program
@@ -80,7 +80,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 		itfTypeCode := b.CreateExtractValue(itf, 0, "")
 		itfValue := b.CreateExtractValue(itf, 1, "")
 		funcPtr = b.getInvokeFunction(&instr.Call)
-		funcType = funcPtr.GlobalValueType()
+		funcPtrType = funcPtr.GlobalValueType()
 		params = append([]llvm.Value{itfValue}, params...) // start with receiver
 		params = append(params, itfTypeCode)               // end with typecode
 	} else {
@@ -90,8 +90,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 		//   * The function context, for closures.
 		//   * The function pointer (for tasks).
 		var context llvm.Value
-		funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value, getPos(instr)))
+		funcPtrType, funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value, getPos(instr)), instr.Call.Value.Type().Underlying().(*types.Signature))
-		funcType = b.getLLVMFunctionType(instr.Call.Value.Type().Underlying().(*types.Signature))
 		params = append(params, context, funcPtr)
 		hasContext = true
 		prefix = b.fn.RelString(nil)
@@ -99,14 +98,14 @@ func (b *builder) createGo(instr *ssa.Go) {
 
 	paramBundle := b.emitPointerPack(params)
 	var stackSize llvm.Value
-	callee := b.createGoroutineStartWrapper(funcType, funcPtr, prefix, hasContext, instr.Pos())
+	callee := b.createGoroutineStartWrapper(funcPtrType, funcPtr, prefix, hasContext, instr.Pos())
 	if b.AutomaticStackSize {
 		// The stack size is not known until after linking. Call a dummy
 		// function that will be replaced with a load from a special ELF
 		// section that contains the stack size (and is modified after
 		// linking).
 		stackSizeFnType, stackSizeFn := b.getFunction(b.program.ImportedPackage("internal/task").Members["getGoroutineStackSize"].(*ssa.Function))
-		stackSize = b.createCall(stackSizeFnType, stackSizeFn, []llvm.Value{callee, llvm.Undef(b.dataPtrType)}, "stacksize")
+		stackSize = b.createCall(stackSizeFnType, stackSizeFn, []llvm.Value{callee, llvm.Undef(b.i8ptrType)}, "stacksize")
 	} else {
 		// The stack size is fixed at compile time. By emitting it here as a
 		// constant, it can be optimized.
@@ -116,7 +115,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 		stackSize = llvm.ConstInt(b.uintptrType, b.DefaultStackSize, false)
 	}
 	fnType, start := b.getFunction(b.program.ImportedPackage("internal/task").Members["start"].(*ssa.Function))
-	b.createCall(fnType, start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.dataPtrType)}, "")
+	b.createCall(fnType, start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.i8ptrType)}, "")
 }
 
 // createGoroutineStartWrapper creates a wrapper for the task-based
@@ -166,7 +165,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 		}
 
 		// Create the wrapper.
-		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.dataPtrType}, false)
+		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
 		wrapper = llvm.AddFunction(c.mod, name+"$gowrapper", wrapperType)
 		c.addStandardAttributes(wrapper)
 		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
@@ -204,7 +203,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 		}
 		params := b.emitPointerUnpack(wrapper.Param(0), paramTypes)
 		if !hasContext {
-			params = append(params, llvm.Undef(c.dataPtrType)) // add dummy context parameter
+			params = append(params, llvm.Undef(c.i8ptrType)) // add dummy context parameter
 		}
 
 		// Create the call.
@@ -212,7 +211,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 
 		if c.Scheduler == "asyncify" {
 			b.CreateCall(deadlockType, deadlock, []llvm.Value{
-				llvm.Undef(c.dataPtrType),
+				llvm.Undef(c.i8ptrType),
 			}, "")
 		}
 
@@ -235,7 +234,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 		// merged into one.
 
 		// Create the wrapper.
-		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.dataPtrType}, false)
+		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
 		wrapper = llvm.AddFunction(c.mod, prefix+".gowrapper", wrapperType)
 		c.addStandardAttributes(wrapper)
 		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
@@ -280,7 +279,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 
 		if c.Scheduler == "asyncify" {
 			b.CreateCall(deadlockType, deadlock, []llvm.Value{
-				llvm.Undef(c.dataPtrType),
+				llvm.Undef(c.i8ptrType),
 			}, "")
 		}
 	}

compiler/inlineasm.go

@@ -99,8 +99,8 @@ func (b *builder) createInlineAsmFull(instr *ssa.CallCommon) (llvm.Value, error)
 			case llvm.IntegerTypeKind:
 				constraints = append(constraints, "r")
 			case llvm.PointerTypeKind:
-				// Memory references require a type starting with LLVM 14,
+				// Memory references require a type in LLVM 14, probably as a
-				// probably as a preparation for opaque pointers.
+				// preparation for opaque pointers.
 				err = b.makeError(instr.Pos(), "support for pointer operands was dropped in TinyGo 0.23")
 				return s
 			default:

compiler/interface.go

@@ -414,10 +414,10 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		}, typeFields...)
 		if hasMethodSet {
 			typeFields = append([]llvm.Value{
-				c.getTypeMethodSet(typ),
+				llvm.ConstBitCast(c.getTypeMethodSet(typ), c.i8ptrType),
 			}, typeFields...)
 		}
-		alignment := c.targetData.TypeAllocSize(c.dataPtrType)
+		alignment := c.targetData.TypeAllocSize(c.i8ptrType)
 		if alignment < 4 {
 			alignment = 4
 		}
@@ -628,7 +628,7 @@ func (c *compilerContext) getTypeMethodSet(typ types.Type) llvm.Value {
 		// Construct global value.
 		globalValue := c.ctx.ConstStruct([]llvm.Value{
 			llvm.ConstInt(c.uintptrType, uint64(ms.Len()), false),
-			llvm.ConstArray(c.dataPtrType, signatures),
+			llvm.ConstArray(c.i8ptrType, signatures),
 			c.ctx.ConstStruct(wrappers, false),
 		}, false)
 		global = llvm.AddGlobal(c.mod, globalValue.Type(), globalName)
@@ -779,7 +779,7 @@ func (c *compilerContext) getInterfaceImplementsFunc(assertedType types.Type) ll
 	fnName := s + ".$typeassert"
 	llvmFn := c.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
-		llvmFnType := llvm.FunctionType(c.ctx.Int1Type(), []llvm.Type{c.dataPtrType}, false)
+		llvmFnType := llvm.FunctionType(c.ctx.Int1Type(), []llvm.Type{c.i8ptrType}, false)
 		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
 		c.addStandardDeclaredAttributes(llvmFn)
 		methods := c.getMethodsString(assertedType.Underlying().(*types.Interface))
@@ -802,7 +802,7 @@ func (c *compilerContext) getInvokeFunction(instr *ssa.CallCommon) llvm.Value {
 			paramTuple = append(paramTuple, sig.Params().At(i))
 		}
 		paramTuple = append(paramTuple, types.NewVar(token.NoPos, nil, "$typecode", types.Typ[types.UnsafePointer]))
-		llvmFnType := c.getLLVMFunctionType(types.NewSignature(sig.Recv(), types.NewTuple(paramTuple...), sig.Results(), false))
+		llvmFnType := c.getRawFuncType(types.NewSignature(sig.Recv(), types.NewTuple(paramTuple...), sig.Results(), false))
 		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
 		c.addStandardDeclaredAttributes(llvmFn)
 		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-invoke", c.getMethodSignatureName(instr.Method)))
@@ -842,7 +842,7 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFnType
 	}
 
 	// create wrapper function
-	paramTypes := append([]llvm.Type{c.dataPtrType}, llvmFnType.ParamTypes()[len(expandedReceiverType):]...)
+	paramTypes := append([]llvm.Type{c.i8ptrType}, llvmFnType.ParamTypes()[len(expandedReceiverType):]...)
 	wrapFnType := llvm.FunctionType(llvmFnType.ReturnType(), paramTypes, false)
 	wrapper = llvm.AddFunction(c.mod, wrapperName, wrapFnType)
 	c.addStandardAttributes(wrapper)

compiler/interrupt.go

@@ -36,7 +36,7 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 		// Fall back to a generic error.
 		return llvm.Value{}, b.makeError(instr.Pos(), "interrupt function must be constant")
 	}
-	funcRawPtr, funcContext := b.decodeFuncValue(funcValue)
+	_, funcRawPtr, funcContext := b.decodeFuncValue(funcValue, nil)
 	funcPtr := llvm.ConstPtrToInt(funcRawPtr, b.uintptrType)
 
 	// Create a new global of type runtime/interrupt.handle. Globals of this

compiler/intrinsics.go

@@ -7,6 +7,7 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )
 
@@ -47,9 +48,12 @@ func (b *builder) defineIntrinsicFunction() {
 func (b *builder) createMemoryCopyImpl() {
 	b.createFunctionStart(true)
 	fnName := "llvm." + b.fn.Name() + ".p0.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+	if llvmutil.Major() < 15 { // compatibility with LLVM 14
+		fnName = "llvm." + b.fn.Name() + ".p0i8.p0i8.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+	}
 	llvmFn := b.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
-		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType, b.dataPtrType, b.uintptrType, b.ctx.Int1Type()}, false)
+		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType, b.i8ptrType, b.uintptrType, b.ctx.Int1Type()}, false)
 		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
 	}
 	var params []llvm.Value
@@ -80,9 +84,12 @@ func (b *builder) createMemoryZeroImpl() {
 // Return the llvm.memset.p0.i8 function declaration.
 func (c *compilerContext) getMemsetFunc() llvm.Value {
 	fnName := "llvm.memset.p0.i" + strconv.Itoa(c.uintptrType.IntTypeWidth())
+	if llvmutil.Major() < 15 { // compatibility with LLVM 14
+		fnName = "llvm.memset.p0i8.i" + strconv.Itoa(c.uintptrType.IntTypeWidth())
+	}
 	llvmFn := c.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
-		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.dataPtrType, c.ctx.Int8Type(), c.uintptrType, c.ctx.Int1Type()}, false)
+		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType, c.ctx.Int8Type(), c.uintptrType, c.ctx.Int1Type()}, false)
 		llvmFn = llvm.AddFunction(c.mod, fnName, fnType)
 	}
 	return llvmFn
@@ -104,7 +111,7 @@ func (b *builder) createKeepAliveImpl() {
 	//
 	// It should be portable to basically everything as the "r" register type
 	// exists basically everywhere.
-	asmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType}, false)
+	asmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType}, false)
 	asmFn := llvm.InlineAsm(asmType, "", "r", true, false, 0, false)
 	b.createCall(asmType, asmFn, []llvm.Value{pointerValue}, "")
 

compiler/llvm.go

@@ -20,7 +20,7 @@ import (
 //
 // 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 (b *builder) createTemporaryAlloca(t llvm.Type, name string) (alloca, size llvm.Value) {
+func (b *builder) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
 	return llvmutil.CreateTemporaryAlloca(b.Builder, b.mod, t, name)
 }
 
@@ -63,45 +63,47 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
 	// Allocate memory for the packed data.
 	size := b.targetData.TypeAllocSize(packedType)
 	if size == 0 {
-		return llvm.ConstPointerNull(b.dataPtrType)
+		return llvm.ConstPointerNull(b.i8ptrType)
 	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
-		return values[0]
+		return b.CreateBitCast(values[0], b.i8ptrType, "pack.ptr")
-	} else if size <= b.targetData.TypeAllocSize(b.dataPtrType) {
+	} else if size <= b.targetData.TypeAllocSize(b.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 b.CreateIntToPtr(values[0], b.dataPtrType, "pack.int")
+			return b.CreateIntToPtr(values[0], b.i8ptrType, "pack.int")
 		}
 
 		// Because packedType is a struct and we have to cast it to a *i8, store
 		// it in a *i8 alloca first and load the *i8 value from there. This is
 		// effectively a bitcast.
-		packedAlloc, _ := b.createTemporaryAlloca(b.dataPtrType, "")
+		packedAlloc, _, _ := b.createTemporaryAlloca(b.i8ptrType, "")
 
-		if size < b.targetData.TypeAllocSize(b.dataPtrType) {
+		if size < b.targetData.TypeAllocSize(b.i8ptrType) {
 			// The alloca is bigger than the value that will be stored in it.
 			// To avoid having some bits undefined, zero the alloca first.
 			// Hopefully this will get optimized away.
-			b.CreateStore(llvm.ConstNull(b.dataPtrType), packedAlloc)
+			b.CreateStore(llvm.ConstNull(b.i8ptrType), packedAlloc)
 		}
 
 		// Store all values in the alloca.
+		packedAllocCast := b.CreateBitCast(packedAlloc, llvm.PointerType(packedType, 0), "")
 		for i, value := range values {
 			indices := []llvm.Value{
 				llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 				llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
 			}
-			gep := b.CreateInBoundsGEP(packedType, packedAlloc, indices, "")
+			gep := b.CreateInBoundsGEP(packedType, packedAllocCast, indices, "")
 			b.CreateStore(value, gep)
 		}
 
 		// Load value (the *i8) from the alloca.
-		result := b.CreateLoad(b.dataPtrType, packedAlloc, "")
+		result := b.CreateLoad(b.i8ptrType, packedAlloc, "")
 
 		// End the lifetime of the alloca, to help the optimizer.
+		packedPtr := b.CreateBitCast(packedAlloc, b.i8ptrType, "")
 		packedSize := llvm.ConstInt(b.ctx.Int64Type(), b.targetData.TypeAllocSize(packedAlloc.Type()), false)
-		b.emitLifetimeEnd(packedAlloc, packedSize)
+		b.emitLifetimeEnd(packedPtr, packedSize)
 
 		return result
 	} else {
@@ -122,20 +124,21 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
 			global.SetGlobalConstant(true)
 			global.SetUnnamedAddr(true)
 			global.SetLinkage(llvm.InternalLinkage)
-			return global
+			return llvm.ConstBitCast(global, b.i8ptrType)
 		}
 
 		// Packed data is bigger than a pointer, so allocate it on the heap.
 		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
 		alloc := b.mod.NamedFunction("runtime.alloc")
-		packedAlloc := b.CreateCall(alloc.GlobalValueType(), alloc, []llvm.Value{
+		packedHeapAlloc := b.CreateCall(alloc.GlobalValueType(), alloc, []llvm.Value{
 			sizeValue,
-			llvm.ConstNull(b.dataPtrType),
+			llvm.ConstNull(b.i8ptrType),
-			llvm.Undef(b.dataPtrType), // unused context parameter
+			llvm.Undef(b.i8ptrType), // unused context parameter
 		}, "")
 		if b.NeedsStackObjects {
-			b.trackPointer(packedAlloc)
+			b.trackPointer(packedHeapAlloc)
 		}
+		packedAlloc := b.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
 
 		// Store all values in the heap pointer.
 		for i, value := range values {
@@ -148,7 +151,7 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
 		}
 
 		// Return the original heap allocation pointer, which already is an *i8.
-		return packedAlloc
+		return packedHeapAlloc
 	}
 }
 
@@ -157,27 +160,28 @@ func (b *builder) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []ll
 	packedType := b.ctx.StructType(valueTypes, false)
 
 	// Get a correctly-typed pointer to the packed data.
-	var packedAlloc llvm.Value
+	var packedAlloc, packedRawAlloc llvm.Value
-	needsLifetimeEnd := false
 	size := b.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{ptr}
+		return []llvm.Value{b.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
-	} else if size <= b.targetData.TypeAllocSize(b.dataPtrType) {
+	} else if size <= b.targetData.TypeAllocSize(b.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{b.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
 		}
 		// Fallback: load it using an alloca.
-		packedAlloc, _ = b.createTemporaryAlloca(b.dataPtrType, "unpack.raw.alloc")
+		packedRawAlloc, _, _ = b.createTemporaryAlloca(llvm.PointerType(b.i8ptrType, 0), "unpack.raw.alloc")
-		b.CreateStore(ptr, packedAlloc)
+		packedRawValue := b.CreateBitCast(ptr, llvm.PointerType(b.i8ptrType, 0), "unpack.raw.value")
-		needsLifetimeEnd = true
+		b.CreateStore(packedRawValue, packedRawAlloc)
+		packedAlloc = b.CreateBitCast(packedRawAlloc, llvm.PointerType(packedType, 0), "unpack.alloc")
 	} else {
-		// Packed data stored on the heap.
+		// Packed data stored on the heap. Bitcast the passed-in pointer to the
-		packedAlloc = ptr
+		// correct pointer type.
+		packedAlloc = b.CreateBitCast(ptr, llvm.PointerType(packedType, 0), "unpack.raw.ptr")
 	}
 	// Load each value from the packed data.
 	values := make([]llvm.Value, len(valueTypes))
@@ -194,9 +198,10 @@ func (b *builder) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []ll
 		gep := b.CreateInBoundsGEP(packedType, packedAlloc, indices, "")
 		values[i] = b.CreateLoad(valueType, gep, "")
 	}
-	if needsLifetimeEnd {
+	if !packedRawAlloc.IsNil() {
+		allocPtr := b.CreateBitCast(packedRawAlloc, b.i8ptrType, "")
 		allocSize := llvm.ConstInt(b.ctx.Int64Type(), b.targetData.TypeAllocSize(b.uintptrType), false)
-		b.emitLifetimeEnd(packedAlloc, allocSize)
+		b.emitLifetimeEnd(allocPtr, allocSize)
 	}
 	return values
 }
@@ -248,12 +253,12 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 	// Do a few checks to see whether we need to generate any object layout
 	// information at all.
 	objectSizeBytes := c.targetData.TypeAllocSize(t)
-	pointerSize := c.targetData.TypeAllocSize(c.dataPtrType)
+	pointerSize := c.targetData.TypeAllocSize(c.i8ptrType)
-	pointerAlignment := c.targetData.PrefTypeAlignment(c.dataPtrType)
+	pointerAlignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
 	if objectSizeBytes < pointerSize {
 		// Too small to contain a pointer.
 		layout := (uint64(1) << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
 	}
 	bitmap := c.getPointerBitmap(t, pos)
 	if bitmap.BitLen() == 0 {
@@ -261,13 +266,13 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		// TODO: this can be done in many other cases, e.g. when allocating an
 		// array (like [4][]byte, which repeats a slice 4 times).
 		layout := (uint64(1) << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
 	}
 	if objectSizeBytes%uint64(pointerAlignment) != 0 {
 		// This shouldn't happen except for packed structs, which aren't
 		// currently used.
 		c.addError(pos, "internal error: unexpected object size for object with pointer field")
-		return llvm.ConstNull(c.dataPtrType)
+		return llvm.ConstNull(c.i8ptrType)
 	}
 	objectSizeWords := objectSizeBytes / uint64(pointerAlignment)
 
@@ -292,7 +297,7 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		// The runtime knows that if the least significant bit of the pointer is
 		// set, the pointer contains the value itself.
 		layout := bitmap.Uint64()<<(sizeFieldBits+1) | (objectSizeWords << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
 	}
 
 	// Unfortunately, the object layout is too big to fit in a pointer-sized
@@ -303,7 +308,7 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 	globalName := "runtime/gc.layout:" + fmt.Sprintf("%d-%0*x", objectSizeWords, (objectSizeWords+15)/16, bitmap)
 	global := c.mod.NamedGlobal(globalName)
 	if !global.IsNil() {
-		return global
+		return llvm.ConstBitCast(global, c.i8ptrType)
 	}
 
 	// Create the global initializer.
@@ -354,13 +359,13 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		global.AddMetadata(0, diglobal)
 	}
 
-	return global
+	return llvm.ConstBitCast(global, c.i8ptrType)
 }
 
 // getPointerBitmap scans the given LLVM type for pointers and sets bits in a
 // bigint at the word offset that contains a pointer. This scan is recursive.
 func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *big.Int {
-	alignment := c.targetData.PrefTypeAlignment(c.dataPtrType)
+	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
 	switch typ.TypeKind() {
 	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
 		return big.NewInt(0)
@@ -373,7 +378,7 @@ func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *big.In
 			// of type uintptr, but before the LowerFuncValues pass it actually
 			// contains a pointer (ptrtoint) to a global. This trips up the
 			// interp package. Therefore, make the id field a pointer for now.
-			typ = c.ctx.StructType([]llvm.Type{c.dataPtrType, c.dataPtrType}, false)
+			typ = c.ctx.StructType([]llvm.Type{c.i8ptrType, c.i8ptrType}, false)
 		}
 		for i, subtyp := range typ.StructElementTypes() {
 			subptrs := c.getPointerBitmap(subtyp, pos)
@@ -453,7 +458,7 @@ func (c *compilerContext) isThumb() bool {
 func (b *builder) readStackPointer() llvm.Value {
 	stacksave := b.mod.NamedFunction("llvm.stacksave")
 	if stacksave.IsNil() {
-		fnType := llvm.FunctionType(b.dataPtrType, nil, false)
+		fnType := llvm.FunctionType(b.i8ptrType, nil, false)
 		stacksave = llvm.AddFunction(b.mod, "llvm.stacksave", fnType)
 	}
 	return b.CreateCall(stacksave.GlobalValueType(), stacksave, nil, "")

compiler/llvmutil/llvm.go

@@ -8,9 +8,22 @@
 package llvmutil
 
 import (
+	"strconv"
+	"strings"
+
 	"tinygo.org/x/go-llvm"
 )
 
+// Major returns the LLVM major version.
+func Major() int {
+	llvmMajor, err := strconv.Atoi(strings.SplitN(llvm.Version, ".", 2)[0])
+	if err != nil {
+		// sanity check, should be unreachable
+		panic("could not parse LLVM version: " + err.Error())
+	}
+	return llvmMajor
+}
+
 // 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.
@@ -33,14 +46,16 @@ func CreateEntryBlockAlloca(builder llvm.Builder, t llvm.Type, name string) llvm
 //
 // This is useful for creating temporary allocas for intrinsics. Don't forget to
 // end the lifetime using emitLifetimeEnd after you're done with it.
-func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, name string) (alloca, size llvm.Value) {
+func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
 	ctx := t.Context()
 	targetData := llvm.NewTargetData(mod.DataLayout())
 	defer targetData.Dispose()
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
 	alloca = CreateEntryBlockAlloca(builder, t, name)
+	bitcast = builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
 	size = llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
 	fnType, fn := getLifetimeStartFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
+	builder.CreateCall(fnType, fn, []llvm.Value{size, bitcast}, "")
 	return
 }
 
@@ -49,19 +64,21 @@ func CreateInstructionAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type,
 	ctx := mod.Context()
 	targetData := llvm.NewTargetData(mod.DataLayout())
 	defer targetData.Dispose()
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
 
 	alloca := CreateEntryBlockAlloca(builder, t, name)
 	builder.SetInsertPointBefore(inst)
+	bitcast := builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
 	size := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
 	fnType, fn := getLifetimeStartFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
+	builder.CreateCall(fnType, fn, []llvm.Value{size, bitcast}, "")
 	if next := llvm.NextInstruction(inst); !next.IsNil() {
 		builder.SetInsertPointBefore(next)
 	} else {
 		builder.SetInsertPointAtEnd(inst.InstructionParent())
 	}
 	fnType, fn = getLifetimeEndFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
+	builder.CreateCall(fnType, fn, []llvm.Value{size, bitcast}, "")
 	return alloca
 }
 
@@ -77,10 +94,13 @@ func EmitLifetimeEnd(builder llvm.Builder, mod llvm.Module, ptr, size llvm.Value
 // first if it doesn't exist yet.
 func getLifetimeStartFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 	fnName := "llvm.lifetime.start.p0"
+	if Major() < 15 { // compatibility with LLVM 14
+		fnName = "llvm.lifetime.start.p0i8"
+	}
 	fn := mod.NamedFunction(fnName)
 	ctx := mod.Context()
-	ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
+	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
 	if fn.IsNil() {
 		fn = llvm.AddFunction(mod, fnName, fnType)
 	}
@@ -91,10 +111,13 @@ func getLifetimeStartFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 // first if it doesn't exist yet.
 func getLifetimeEndFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 	fnName := "llvm.lifetime.end.p0"
+	if Major() < 15 {
+		fnName = "llvm.lifetime.end.p0i8"
+	}
 	fn := mod.NamedFunction(fnName)
 	ctx := mod.Context()
-	ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
+	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
 	if fn.IsNil() {
 		fn = llvm.AddFunction(mod, fnName, fnType)
 	}
@@ -190,15 +213,13 @@ func AppendToGlobal(mod llvm.Module, globalName string, values ...llvm.Value) {
 	}
 
 	// Add the new values.
-	ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
 	for _, value := range values {
-		// Note: the bitcast is necessary to cast AVR function pointers to
+		usedValues = append(usedValues, llvm.ConstPointerCast(value, i8ptrType))
-		// address space 0 pointer types.
-		usedValues = append(usedValues, llvm.ConstPointerCast(value, ptrType))
 	}
 
 	// Create a new array (with the old and new values).
-	usedInitializer := llvm.ConstArray(ptrType, usedValues)
+	usedInitializer := llvm.ConstArray(i8ptrType, usedValues)
 	used := llvm.AddGlobal(mod, usedInitializer.Type(), globalName)
 	used.SetInitializer(usedInitializer)
 	used.SetLinkage(llvm.AppendingLinkage)

compiler/map.go

@@ -65,7 +65,7 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 	// 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, mapValueAllocaSize := b.createTemporaryAlloca(llvmValueType, "hashmap.value")
+	mapValueAlloca, mapValuePtr, mapValueAllocaSize := b.createTemporaryAlloca(llvmValueType, "hashmap.value")
 
 	// We need the map size (with type uintptr) to pass to the hashmap*Get
 	// functions. This is necessary because those *Get functions are valid on
@@ -82,19 +82,19 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, mapValueAlloca, mapValueSize}
+		params := []llvm.Value{m, key, mapValuePtr, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapStringGet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
 		// Store the key in an alloca, in the entry block to avoid dynamic stack
 		// growth.
-		mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		mapKeyAlloca, mapKeyPtr, mapKeySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, mapKeyAlloca)
 		b.zeroUndefBytes(b.getLLVMType(keyType), mapKeyAlloca)
 		// Fetch the value from the hashmap.
-		params := []llvm.Value{m, mapKeyAlloca, mapValueAlloca, mapValueSize}
+		params := []llvm.Value{m, mapKeyPtr, mapValuePtr, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapBinaryGet", params, "")
-		b.emitLifetimeEnd(mapKeyAlloca, mapKeySize)
+		b.emitLifetimeEnd(mapKeyPtr, mapKeySize)
 	} else {
 		// Not trivially comparable using memcmp. Make it an interface instead.
 		itfKey := key
@@ -102,14 +102,14 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 			// Not already an interface, so convert it to an interface now.
 			itfKey = b.createMakeInterface(key, origKeyType, pos)
 		}
-		params := []llvm.Value{m, itfKey, mapValueAlloca, mapValueSize}
+		params := []llvm.Value{m, itfKey, mapValuePtr, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapInterfaceGet", params, "")
 	}
 
 	// Load the resulting value from the hashmap. The value is set to the zero
 	// value if the key doesn't exist in the hashmap.
 	mapValue := b.CreateLoad(llvmValueType, mapValueAlloca, "")
-	b.emitLifetimeEnd(mapValueAlloca, mapValueAllocaSize)
+	b.emitLifetimeEnd(mapValuePtr, mapValueAllocaSize)
 
 	if commaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{llvmValueType, b.ctx.Int1Type()}, false))
@@ -124,22 +124,22 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 // createMapUpdate updates a map key to a given value, by creating an
 // appropriate runtime call.
 func (b *builder) createMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
-	valueAlloca, valueSize := b.createTemporaryAlloca(value.Type(), "hashmap.value")
+	valueAlloca, valuePtr, valueSize := b.createTemporaryAlloca(value.Type(), "hashmap.value")
 	b.CreateStore(value, valueAlloca)
 	origKeyType := keyType
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, valueAlloca}
+		params := []llvm.Value{m, key, valuePtr}
 		b.createRuntimeCall("hashmapStringSet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keyPtr, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, keyAlloca)
 		b.zeroUndefBytes(b.getLLVMType(keyType), keyAlloca)
-		params := []llvm.Value{m, keyAlloca, valueAlloca}
+		params := []llvm.Value{m, keyPtr, valuePtr}
 		b.createRuntimeCall("hashmapBinarySet", params, "")
-		b.emitLifetimeEnd(keyAlloca, keySize)
+		b.emitLifetimeEnd(keyPtr, keySize)
 	} else {
 		// Key is not trivially comparable, so compare it as an interface instead.
 		itfKey := key
@@ -147,10 +147,10 @@ func (b *builder) createMapUpdate(keyType types.Type, m, key, value llvm.Value,
 			// Not already an interface, so convert it to an interface first.
 			itfKey = b.createMakeInterface(key, origKeyType, pos)
 		}
-		params := []llvm.Value{m, itfKey, valueAlloca}
+		params := []llvm.Value{m, itfKey, valuePtr}
 		b.createRuntimeCall("hashmapInterfaceSet", params, "")
 	}
-	b.emitLifetimeEnd(valueAlloca, valueSize)
+	b.emitLifetimeEnd(valuePtr, valueSize)
 }
 
 // createMapDelete deletes a key from a map by calling the appropriate runtime
@@ -164,12 +164,12 @@ func (b *builder) createMapDelete(keyType types.Type, m, key llvm.Value, pos tok
 		b.createRuntimeCall("hashmapStringDelete", params, "")
 		return nil
 	} else if hashmapIsBinaryKey(keyType) {
-		keyAlloca, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keyPtr, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, keyAlloca)
 		b.zeroUndefBytes(b.getLLVMType(keyType), keyAlloca)
-		params := []llvm.Value{m, keyAlloca}
+		params := []llvm.Value{m, keyPtr}
 		b.createRuntimeCall("hashmapBinaryDelete", params, "")
-		b.emitLifetimeEnd(keyAlloca, keySize)
+		b.emitLifetimeEnd(keyPtr, keySize)
 		return nil
 	} else {
 		// Key is not trivially comparable, so compare it as an interface
@@ -225,9 +225,9 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 	}
 
 	// Extract the key and value from the map.
-	mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(llvmStoredKeyType, "range.key")
+	mapKeyAlloca, mapKeyPtr, mapKeySize := b.createTemporaryAlloca(llvmStoredKeyType, "range.key")
-	mapValueAlloca, mapValueSize := b.createTemporaryAlloca(llvmValueType, "range.value")
+	mapValueAlloca, mapValuePtr, mapValueSize := b.createTemporaryAlloca(llvmValueType, "range.value")
-	ok := b.createRuntimeCall("hashmapNext", []llvm.Value{llvmRangeVal, it, mapKeyAlloca, mapValueAlloca}, "range.next")
+	ok := b.createRuntimeCall("hashmapNext", []llvm.Value{llvmRangeVal, it, mapKeyPtr, mapValuePtr}, "range.next")
 	mapKey := b.CreateLoad(llvmStoredKeyType, mapKeyAlloca, "")
 	mapValue := b.CreateLoad(llvmValueType, mapValueAlloca, "")
 
@@ -238,8 +238,8 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 	}
 
 	// End the lifetimes of the allocas, because we're done with them.
-	b.emitLifetimeEnd(mapKeyAlloca, mapKeySize)
+	b.emitLifetimeEnd(mapKeyPtr, mapKeySize)
-	b.emitLifetimeEnd(mapValueAlloca, mapValueSize)
+	b.emitLifetimeEnd(mapValuePtr, mapValueSize)
 
 	// Construct the *ssa.Next return value: {ok, mapKey, mapValue}
 	tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{b.ctx.Int1Type(), llvmKeyType, llvmValueType}, false))
@@ -333,7 +333,14 @@ func (b *builder) zeroUndefBytes(llvmType llvm.Type, ptr llvm.Value) error {
 			if fieldEndOffset != nextOffset {
 				n := llvm.ConstInt(b.uintptrType, nextOffset-fieldEndOffset, false)
 				llvmStoreSize := llvm.ConstInt(b.uintptrType, storeSize, false)
-				paddingStart := b.CreateInBoundsGEP(b.ctx.Int8Type(), elemPtr, []llvm.Value{llvmStoreSize}, "")
+				gepPtr := elemPtr
+				if gepPtr.Type() != b.i8ptrType {
+					gepPtr = b.CreateBitCast(gepPtr, b.i8ptrType, "") // LLVM 14
+				}
+				paddingStart := b.CreateInBoundsGEP(b.ctx.Int8Type(), gepPtr, []llvm.Value{llvmStoreSize}, "")
+				if paddingStart.Type() != b.i8ptrType {
+					paddingStart = b.CreateBitCast(paddingStart, b.i8ptrType, "") // LLVM 14
+				}
 				b.createRuntimeCall("memzero", []llvm.Value{paddingStart, n}, "")
 			}
 		}

compiler/symbol.go

@@ -96,7 +96,7 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	// Add an extra parameter as the function context. This context is used in
 	// closures and bound methods, but should be optimized away when not used.
 	if !info.exported {
-		paramInfos = append(paramInfos, paramInfo{llvmType: c.dataPtrType, name: "context", elemSize: 0})
+		paramInfos = append(paramInfos, paramInfo{llvmType: c.i8ptrType, name: "context", elemSize: 0})
 	}
 
 	var paramTypes []llvm.Type
@@ -145,18 +145,20 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		for _, attrName := range []string{"noalias", "nonnull"} {
 			llvmFn.AddAttributeAtIndex(0, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(attrName), 0))
 		}
-		// Add attributes to signal to LLVM that this is an allocator function.
+		if llvmutil.Major() >= 15 { // allockind etc are not available in LLVM 14
-		// This enables a number of optimizations.
+			// Add attributes to signal to LLVM that this is an allocator
-		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allockind"), allocKindAlloc|allocKindZeroed))
+			// function. This enables a number of optimizations.
-		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("alloc-family", "runtime.alloc"))
+			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allockind"), allocKindAlloc|allocKindZeroed))
-		// Use a special value to indicate the first parameter:
+			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("alloc-family", "runtime.alloc"))
-		// > allocsize has two integer arguments, but because they're both 32 bits, we can
+			// Use a special value to indicate the first parameter:
-		// > pack them into one 64-bit value, at the cost of making said value
+			// > allocsize has two integer arguments, but because they're both 32 bits, we can
-		// > nonsensical.
+			// > pack them into one 64-bit value, at the cost of making said value
-		// >
+			// > nonsensical.
-		// > In order to do this, we need to reserve one value of the second (optional)
+			// >
-		// > allocsize argument to signify "not present."
+			// > In order to do this, we need to reserve one value of the second (optional)
-		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allocsize"), 0x0000_0000_ffff_ffff))
+			// > allocsize argument to signify "not present."
+			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allocsize"), 0x0000_0000_ffff_ffff))
+		}
 	case "runtime.sliceAppend":
 		// Appending a slice will only read the to-be-appended slice, it won't
 		// be modified.
@@ -443,10 +445,15 @@ func (c *compilerContext) addStandardDefinedAttributes(llvmFn llvm.Value) {
 	llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nounwind"), 0))
 	if strings.Split(c.Triple, "-")[0] == "x86_64" {
 		// Required by the ABI.
-		// The uwtable has two possible values: sync (1) or async (2). We use
+		if llvmutil.Major() < 15 {
-		// sync because we currently don't use async unwind tables.
+			// Needed for LLVM 14 support.
-		// For details, see: https://llvm.org/docs/LangRef.html#function-attributes
+			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 0))
-		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 1))
+		} else {
+			// The uwtable has two possible values: sync (1) or async (2). We
+			// use sync because we currently don't use async unwind tables.
+			// For details, see: https://llvm.org/docs/LangRef.html#function-attributes
+			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 1))
+		}
 	}
 }
 

compiler/syscall.go

@@ -183,7 +183,7 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		}
 		llvmType := llvm.FunctionType(b.uintptrType, paramTypes, false)
 		fn := b.getValue(call.Args[0], getPos(call))
-		fnPtr := b.CreateIntToPtr(fn, b.dataPtrType, "")
+		fnPtr := b.CreateIntToPtr(fn, llvm.PointerType(llvmType, 0), "")
 
 		// Prepare some functions that will be called later.
 		setLastError := b.mod.NamedFunction("SetLastError")

docs/Makefile

@@ -16,7 +16,7 @@ I18NSPHINXOPTS  = $(PAPEROPT_$(PAPER)) $(SPHINXOPTS) .
 
 .PHONY: help
 help:
-	@echo "Please use \`$(MAKE) <target>' where <target> is one of"
+	@echo "Please use \`make <target>' where <target> is one of"
 	@echo "  html       to make standalone HTML files"
 	@echo "  dirhtml    to make HTML files named index.html in directories"
 	@echo "  singlehtml to make a single large HTML file"

go.mod

@@ -18,7 +18,7 @@ require (
 	golang.org/x/sys v0.11.0
 	golang.org/x/tools v0.12.0
 	gopkg.in/yaml.v2 v2.4.0
-	tinygo.org/x/go-llvm v0.0.0-20230920233244-32ed56c6be9c
+	tinygo.org/x/go-llvm v0.0.0-20230918183930-9edb6403d0bc
 )
 
 require (

go.sum

@@ -65,5 +65,5 @@ gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8
 gopkg.in/yaml.v2 v2.4.0 h1:D8xgwECY7CYvx+Y2n4sBz93Jn9JRvxdiyyo8CTfuKaY=
 gopkg.in/yaml.v2 v2.4.0/go.mod h1:RDklbk79AGWmwhnvt/jBztapEOGDOx6ZbXqjP6csGnQ=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c h1:dUUwHk2QECo/6vqA44rthZ8ie2QXMNeKRTHCNY2nXvo=
-tinygo.org/x/go-llvm v0.0.0-20230920233244-32ed56c6be9c h1:rS8mAFqf0CfxPCbtfsI3bWL4jkb0TBYA1wx7tY1nu28=
+tinygo.org/x/go-llvm v0.0.0-20230918183930-9edb6403d0bc h1:IVX1dqCX3c88P7iEMBtz1xCAM4UIqCMgbqHdSefBaWE=
-tinygo.org/x/go-llvm v0.0.0-20230920233244-32ed56c6be9c/go.mod h1:GFbusT2VTA4I+l4j80b17KFK+6whv69Wtny5U+T8RR0=
+tinygo.org/x/go-llvm v0.0.0-20230918183930-9edb6403d0bc/go.mod h1:GFbusT2VTA4I+l4j80b17KFK+6whv69Wtny5U+T8RR0=

interp/errors.go

@@ -60,10 +60,10 @@ func (r *runner) errorAt(inst instruction, err error) *Error {
 	pos := getPosition(inst.llvmInst)
 	return &Error{
 		ImportPath: r.pkgName,
-		Inst:       inst.llvmInst.String(),
+		Inst:       inst.String(),
 		Pos:        pos,
 		Err:        err,
-		Traceback:  []ErrorLine{{pos, inst.llvmInst.String()}},
+		Traceback:  []ErrorLine{{pos, inst.String()}},
 	}
 }
 

interp/interp.go

@@ -21,7 +21,7 @@ type runner struct {
 	targetData    llvm.TargetData
 	builder       llvm.Builder
 	pointerSize   uint32                   // cached pointer size from the TargetData
-	dataPtrType   llvm.Type                // often used type so created in advance
+	i8ptrType     llvm.Type                // often used type so created in advance
 	uintptrType   llvm.Type                // equivalent to uintptr in Go
 	maxAlign      int                      // maximum alignment of an object, alignment of runtime.alloc() result
 	debug         bool                     // log debug messages
@@ -46,9 +46,9 @@ func newRunner(mod llvm.Module, timeout time.Duration, debug bool) *runner {
 		timeout:       timeout,
 	}
 	r.pointerSize = uint32(r.targetData.PointerSize())
-	r.dataPtrType = llvm.PointerType(mod.Context().Int8Type(), 0)
+	r.i8ptrType = llvm.PointerType(mod.Context().Int8Type(), 0)
 	r.uintptrType = mod.Context().IntType(r.targetData.PointerSize() * 8)
-	r.maxAlign = r.targetData.PrefTypeAlignment(r.dataPtrType) // assume pointers are maximally aligned (this is not always the case)
+	r.maxAlign = r.targetData.PrefTypeAlignment(r.i8ptrType) // assume pointers are maximally aligned (this is not always the case)
 	return &r
 }
 
@@ -126,7 +126,7 @@ func Run(mod llvm.Module, timeout time.Duration, debug bool) error {
 				mem.revert()
 				// Create a call to the package initializer (which was
 				// previously deleted).
-				i8undef := llvm.Undef(r.dataPtrType)
+				i8undef := llvm.Undef(r.i8ptrType)
 				r.builder.CreateCall(fn.GlobalValueType(), fn, []llvm.Value{i8undef}, "")
 				// Make sure that any globals touched by the package
 				// initializer, won't be accessed by later package initializers.
@@ -174,7 +174,8 @@ func Run(mod llvm.Module, timeout time.Duration, debug bool) error {
 			newGlobal.SetLinkage(obj.llvmGlobal.Linkage())
 			newGlobal.SetAlignment(obj.llvmGlobal.Alignment())
 			// TODO: copy debug info, unnamed_addr, ...
-			obj.llvmGlobal.ReplaceAllUsesWith(newGlobal)
+			bitcast := llvm.ConstBitCast(newGlobal, obj.llvmGlobal.Type())
+			obj.llvmGlobal.ReplaceAllUsesWith(bitcast)
 			name := obj.llvmGlobal.Name()
 			obj.llvmGlobal.EraseFromParentAsGlobal()
 			newGlobal.SetName(name)

interp/interp_test.go

@@ -77,9 +77,12 @@ func runTest(t *testing.T, pathPrefix string) {
 	}
 
 	// Run some cleanup passes to get easy-to-read outputs.
-	to := llvm.NewPassBuilderOptions()
+	pm := llvm.NewPassManager()
-	defer to.Dispose()
+	defer pm.Dispose()
-	mod.RunPasses("globalopt,dse,adce", llvm.TargetMachine{}, to)
+	pm.AddGlobalOptimizerPass()
+	pm.AddDeadStoreEliminationPass()
+	pm.AddAggressiveDCEPass()
+	pm.Run(mod)
 
 	// Read the expected output IR.
 	out, err := os.ReadFile(pathPrefix + ".out.ll")

interp/interpreter.go

@@ -543,7 +543,7 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					// how this function got called.
 					callErr.Traceback = append(callErr.Traceback, ErrorLine{
 						Pos:  getPosition(inst.llvmInst),
-						Inst: inst.llvmInst.String(),
+						Inst: inst.String(),
 					})
 					return nil, mem, callErr
 				}
@@ -1046,3 +1046,15 @@ func intPredicateString(predicate llvm.IntPredicate) string {
 		return "cmp?"
 	}
 }
+
+// Strip some pointer casts. This is probably unnecessary once support for
+// LLVM 14 (non-opaque pointers) is dropped.
+func stripPointerCasts(value llvm.Value) llvm.Value {
+	if !value.IsAConstantExpr().IsNil() {
+		switch value.Opcode() {
+		case llvm.GetElementPtr, llvm.BitCast:
+			return stripPointerCasts(value.Operand(0))
+		}
+	}
+	return value
+}

interp/memory.go

@@ -658,11 +658,20 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 	if v.offset() != 0 {
 		// If there is an offset, make sure to use a GEP to index into the
 		// pointer.
+		// Cast to an i8* first (if needed) for easy indexing.
+		if llvmValue.Type() != mem.r.i8ptrType {
+			llvmValue = llvm.ConstBitCast(llvmValue, mem.r.i8ptrType)
+		}
 		llvmValue = llvm.ConstInBoundsGEP(mem.r.mod.Context().Int8Type(), llvmValue, []llvm.Value{
 			llvm.ConstInt(mem.r.mod.Context().Int32Type(), uint64(v.offset()), false),
 		})
 	}
 
+	// If a particular LLVM pointer type is requested, cast to it.
+	if !llvmType.IsNil() && llvmType != llvmValue.Type() {
+		llvmValue = llvm.ConstBitCast(llvmValue, llvmType)
+	}
+
 	return llvmValue, nil
 }
 
@@ -863,7 +872,7 @@ func (v rawValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Value,
 			if err != nil {
 				panic(err)
 			}
-			if checks && mem.r.targetData.TypeAllocSize(llvmType) != mem.r.targetData.TypeAllocSize(mem.r.dataPtrType) {
+			if checks && mem.r.targetData.TypeAllocSize(llvmType) != mem.r.targetData.TypeAllocSize(mem.r.i8ptrType) {
 				// Probably trying to serialize a pointer to a byte array,
 				// perhaps as a result of rawLLVMValue() in a previous interp
 				// run.
@@ -945,6 +954,8 @@ func (v rawValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Value,
 					// Because go-llvm doesn't have addrspacecast at the moment,
 					// do it indirectly with a ptrtoint/inttoptr pair.
 					llvmValue = llvm.ConstIntToPtr(llvm.ConstPtrToInt(llvmValue, mem.r.uintptrType), llvmType)
+				} else {
+					llvmValue = llvm.ConstBitCast(llvmValue, llvmType)
 				}
 			}
 			return llvmValue, nil
@@ -1245,7 +1256,7 @@ func (r *runner) getValue(llvmValue llvm.Value) value {
 // For details on this format, see src/runtime/gc_precise.go.
 func (r *runner) readObjectLayout(layoutValue value) (uint64, *big.Int) {
 	pointerSize := layoutValue.len(r)
-	if checks && uint64(pointerSize) != r.targetData.TypeAllocSize(r.dataPtrType) {
+	if checks && uint64(pointerSize) != r.targetData.TypeAllocSize(r.i8ptrType) {
 		panic("inconsistent pointer size")
 	}
 
@@ -1320,12 +1331,12 @@ func (r *runner) getLLVMTypeFromLayout(layoutValue value) llvm.Type {
 
 	// Create the LLVM type.
 	pointerSize := layoutValue.len(r)
-	pointerAlignment := r.targetData.PrefTypeAlignment(r.dataPtrType)
+	pointerAlignment := r.targetData.PrefTypeAlignment(r.i8ptrType)
 	var fields []llvm.Type
 	for i := 0; i < int(objectSizeWords); {
 		if bitmap.Bit(i) != 0 {
 			// Pointer field.
-			fields = append(fields, r.dataPtrType)
+			fields = append(fields, r.i8ptrType)
 			i += int(pointerSize / uint32(pointerAlignment))
 		} else {
 			// Byte/word field.

targets/avrtiny.S

@@ -1,3 +1,4 @@
+; TODO: remove these in LLVM 15
 #define __tmp_reg__ r16
 #define __zero_reg__ r17
 

testdata/cgo/main.go

@@ -9,7 +9,6 @@ int mul(int, int);
 #include <string.h>
 #cgo CFLAGS: -DSOME_CONSTANT=17
 #define someDefine -5 + 2 * 7
-bool someBool;
 */
 import "C"
 
@@ -53,13 +52,6 @@ func main() {
 	println("static headerfunc:", C.headerfunc_static(5))
 	headerfunc_2()
 
-	// equivalent types
-	var goInt8 int8 = 5
-	var _ C.int8_t = goInt8
-
-	var _ bool = C.someBool
-	var _ C._Bool = C.someBool
-
 	// more globals
 	println("bool:", C.globalBool, C.globalBool2 == true)
 	println("float:", C.globalFloat)

tools/gen-device-svd/gen-device-svd.go

@@ -10,11 +10,9 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
-	"runtime"
 	"sort"
 	"strconv"
 	"strings"
-	"sync"
 	"text/template"
 	"unicode"
 )
@@ -1443,31 +1441,6 @@ __isr_vector:
 	return w.Flush()
 }
 
-// Process a single SVD file, synchronously.
-func processFile(infile, outdir, sourceURL, interruptSystem string) error {
-	device, err := readSVD(infile, sourceURL)
-	if err != nil {
-		return fmt.Errorf("failed to read: %w", err)
-	}
-	err = writeGo(outdir, device, interruptSystem)
-	if err != nil {
-		return fmt.Errorf("failed to write Go file: %w", err)
-	}
-	switch interruptSystem {
-	case "software":
-		// Nothing to do.
-	case "hardware":
-		err = writeAsm(outdir, device)
-		if err != nil {
-			return fmt.Errorf("failed to write assembly file: %w", err)
-		}
-	default:
-		return fmt.Errorf("unknown interrupt system: %s", interruptSystem)
-	}
-	return nil
-}
-
-// Process an entire directory, in parallel.
 func generate(indir, outdir, sourceURL, interruptSystem string) error {
 	if _, err := os.Stat(indir); errors.Is(err, fs.ErrNotExist) {
 		fmt.Fprintln(os.Stderr, "cannot find input directory:", indir)
@@ -1475,52 +1448,35 @@ func generate(indir, outdir, sourceURL, interruptSystem string) error {
 	}
 	os.MkdirAll(outdir, 0777)
 
-	// Read list of SVD files to process.
 	infiles, err := filepath.Glob(filepath.Join(indir, "*.svd"))
 	if err != nil {
 		fmt.Fprintln(os.Stderr, "could not read .svd files:", err)
 		os.Exit(1)
 	}
 	sort.Strings(infiles)
-
+	for _, infile := range infiles {
-	// Start worker goroutines.
+		fmt.Println(infile)
-	var wg sync.WaitGroup
+		device, err := readSVD(infile, sourceURL)
-	workChan := make(chan string)
+		if err != nil {
-	errChan := make(chan error, 1)
+			return fmt.Errorf("failed to read: %w", err)
-	for i := 0; i < runtime.NumCPU(); i++ {
+		}
-		go func() {
+		err = writeGo(outdir, device, interruptSystem)
-			for infile := range workChan {
+		if err != nil {
-				err := processFile(infile, outdir, sourceURL, interruptSystem)
+			return fmt.Errorf("failed to write Go file: %w", err)
-				if err != nil {
+		}
-					// Store error to errChan if no error was stored before.
+		switch interruptSystem {
-					select {
+		case "software":
-					case errChan <- err:
+			// Nothing to do.
-					default:
+		case "hardware":
-					}
+			err = writeAsm(outdir, device)
-				}
+			if err != nil {
-				wg.Done()
+				return fmt.Errorf("failed to write assembly file: %w", err)
 			}
-		}()
+		default:
-	}
+			return fmt.Errorf("unknown interrupt system: %s", interruptSystem)
-
+		}
-	// Submit all jobs to the goroutines.
-	wg.Add(len(infiles))
-	for _, filepath := range infiles {
-		fmt.Println(filepath)
-		workChan <- filepath
-	}
-	close(workChan)
-
-	// Wait until all workers have finished.
-	wg.Wait()
-
-	// Check for an error.
-	select {
-	case err := <-errChan:
-		return err
-	default:
-		return nil
 	}
+	return nil
 }
 
 func main() {

transform/allocs.go

@@ -37,7 +37,7 @@ func OptimizeAllocs(mod llvm.Module, printAllocs *regexp.Regexp, logger func(tok
 
 	targetData := llvm.NewTargetData(mod.DataLayout())
 	defer targetData.Dispose()
-	ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
 	builder := mod.Context().NewBuilder()
 	defer builder.Dispose()
 
@@ -110,7 +110,7 @@ func OptimizeAllocs(mod llvm.Module, printAllocs *regexp.Regexp, logger func(tok
 		} else {
 			alignment = 8
 		}
-		if pointerAlignment := targetData.ABITypeAlignment(ptrType); pointerAlignment < alignment {
+		if pointerAlignment := targetData.ABITypeAlignment(i8ptrType); pointerAlignment < alignment {
 			// Use min(alignment, alignof(void*)) as the alignment.
 			alignment = pointerAlignment
 		}
@@ -120,7 +120,7 @@ func OptimizeAllocs(mod llvm.Module, printAllocs *regexp.Regexp, logger func(tok
 		fn := bitcast.InstructionParent().Parent()
 		builder.SetInsertPointBefore(fn.EntryBasicBlock().FirstInstruction())
 		allocaType := llvm.ArrayType(mod.Context().Int8Type(), int(size))
-		alloca := builder.CreateAlloca(allocaType, "stackalloc")
+		alloca := builder.CreateAlloca(allocaType, "stackalloc.alloca")
 		alloca.SetAlignment(alignment)
 
 		// Zero the allocation inside the block where the value was originally allocated.
@@ -130,7 +130,8 @@ func OptimizeAllocs(mod llvm.Module, printAllocs *regexp.Regexp, logger func(tok
 		store.SetAlignment(alignment)
 
 		// Replace heap alloc bitcast with stack alloc bitcast.
-		bitcast.ReplaceAllUsesWith(alloca)
+		stackalloc := builder.CreateBitCast(alloca, bitcast.Type(), "stackalloc")
+		bitcast.ReplaceAllUsesWith(stackalloc)
 		if heapalloc != bitcast {
 			bitcast.EraseFromParentAsInstruction()
 		}

transform/allocs_test.go

@@ -38,12 +38,11 @@ func TestAllocs2(t *testing.T) {
 	mod := compileGoFileForTesting(t, "./testdata/allocs2.go")
 
 	// Run functionattrs pass, which is necessary for escape analysis.
-	po := llvm.NewPassBuilderOptions()
+	pm := llvm.NewPassManager()
-	defer po.Dispose()
+	defer pm.Dispose()
-	err := mod.RunPasses("function(instcombine),function-attrs", llvm.TargetMachine{}, po)
+	pm.AddInstructionCombiningPass()
-	if err != nil {
+	pm.AddFunctionAttrsPass()
-		t.Error("failed to run passes:", err)
+	pm.Run(mod)
-	}
 
 	// Run heap to stack transform.
 	var testOutputs []allocsTestOutput

transform/gc.go

@@ -225,7 +225,8 @@ func MakeGCStackSlots(mod llvm.Module) bool {
 			llvm.ConstInt(ctx.Int32Type(), 0, false),
 		}, "")
 		builder.CreateStore(parent, gep)
-		builder.CreateStore(stackObject, stackChainStart)
+		stackObjectCast := builder.CreateBitCast(stackObject, stackChainStartType, "")
+		builder.CreateStore(stackObjectCast, stackChainStart)
 
 		// Do a store to the stack object after each new pointer that is created.
 		pointerStores := make(map[llvm.Value]struct{})

transform/interface-lowering.go

@@ -92,7 +92,7 @@ type lowerInterfacesPass struct {
 	ctx         llvm.Context
 	uintptrType llvm.Type
 	targetData  llvm.TargetData
-	ptrType     llvm.Type
+	i8ptrType   llvm.Type
 	types       map[string]*typeInfo
 	signatures  map[string]*signatureInfo
 	interfaces  map[string]*interfaceInfo
@@ -113,7 +113,7 @@ func LowerInterfaces(mod llvm.Module, config *compileopts.Config) error {
 		ctx:         ctx,
 		targetData:  targetData,
 		uintptrType: mod.Context().IntType(targetData.PointerSize() * 8),
-		ptrType:     llvm.PointerType(ctx.Int8Type(), 0),
+		i8ptrType:   llvm.PointerType(ctx.Int8Type(), 0),
 		types:       make(map[string]*typeInfo),
 		signatures:  make(map[string]*signatureInfo),
 		interfaces:  make(map[string]*interfaceInfo),
@@ -356,9 +356,11 @@ func (p *lowerInterfacesPass) run() error {
 			}
 			// Fallback.
 			if hasUses(t.typecode) {
-				negativeOffset := -int64(p.targetData.TypeAllocSize(p.ptrType))
+				bitcast := llvm.ConstBitCast(newGlobal, p.i8ptrType)
-				gep := p.builder.CreateInBoundsGEP(p.ctx.Int8Type(), newGlobal, []llvm.Value{llvm.ConstInt(p.ctx.Int32Type(), uint64(negativeOffset), true)}, "")
+				negativeOffset := -int64(p.targetData.TypeAllocSize(p.i8ptrType))
-				t.typecode.ReplaceAllUsesWith(gep)
+				gep := p.builder.CreateInBoundsGEP(p.ctx.Int8Type(), bitcast, []llvm.Value{llvm.ConstInt(p.ctx.Int32Type(), uint64(negativeOffset), true)}, "")
+				bitcast2 := llvm.ConstBitCast(gep, t.typecode.Type())
+				t.typecode.ReplaceAllUsesWith(bitcast2)
 			}
 			t.typecode.EraseFromParentAsGlobal()
 			newGlobal.SetName(typecodeName)
@@ -512,7 +514,7 @@ func (p *lowerInterfacesPass) defineInterfaceMethodFunc(fn llvm.Value, itf *inte
 		params[i] = fn.Param(i + 1)
 	}
 	params = append(params,
-		llvm.Undef(p.ptrType),
+		llvm.Undef(p.i8ptrType),
 	)
 
 	// Start chain in the entry block.
@@ -552,12 +554,27 @@ func (p *lowerInterfacesPass) defineInterfaceMethodFunc(fn llvm.Value, itf *inte
 
 		p.builder.SetInsertPointAtEnd(bb)
 		receiver := fn.FirstParam()
+		if receiver.Type() != function.FirstParam().Type() {
+			// When the receiver is a pointer, it is not wrapped. This means the
+			// i8* has to be cast to the correct pointer type of the target
+			// function.
+			receiver = p.builder.CreateBitCast(receiver, function.FirstParam().Type(), "")
+		}
 
+		// Check whether the called function has the same signature as would be
+		// expected from the parameters. This can happen in rare cases when
+		// named struct types are renamed after merging multiple LLVM modules.
 		paramTypes := []llvm.Type{receiver.Type()}
 		for _, param := range params {
 			paramTypes = append(paramTypes, param.Type())
 		}
+		calledFunctionType := function.Type()
 		functionType := llvm.FunctionType(returnType, paramTypes, false)
+		sig := llvm.PointerType(functionType, calledFunctionType.PointerAddressSpace())
+		if sig != function.Type() {
+			function = p.builder.CreateBitCast(function, sig, "")
+		}
+
 		retval := p.builder.CreateCall(functionType, function, append([]llvm.Value{receiver}, params...), "")
 		if retval.Type().TypeKind() == llvm.VoidTypeKind {
 			p.builder.CreateRetVoid()
@@ -579,7 +596,7 @@ func (p *lowerInterfacesPass) defineInterfaceMethodFunc(fn llvm.Value, itf *inte
 	// method on a nil interface.
 	nilPanic := p.mod.NamedFunction("runtime.nilPanic")
 	p.builder.CreateCall(nilPanic.GlobalValueType(), nilPanic, []llvm.Value{
-		llvm.Undef(p.ptrType),
+		llvm.Undef(p.i8ptrType),
 	}, "")
 	p.builder.CreateUnreachable()
 }

transform/interface-lowering_test.go

@@ -15,11 +15,9 @@ func TestInterfaceLowering(t *testing.T) {
 			t.Error(err)
 		}
 
-		po := llvm.NewPassBuilderOptions()
+		pm := llvm.NewPassManager()
-		defer po.Dispose()
+		defer pm.Dispose()
-		err = mod.RunPasses("globaldce", llvm.TargetMachine{}, po)
+		pm.AddGlobalDCEPass()
-		if err != nil {
+		pm.Run(mod)
-			t.Error("failed to run passes:", err)
-		}
 	})
 }

transform/maps_test.go

@@ -15,11 +15,10 @@ func TestOptimizeMaps(t *testing.T) {
 
 		// Run an optimization pass, to clean up the result.
 		// This shows that all code related to the map is really eliminated.
-		po := llvm.NewPassBuilderOptions()
+		pm := llvm.NewPassManager()
-		defer po.Dispose()
+		defer pm.Dispose()
-		err := mod.RunPasses("dse,adce", llvm.TargetMachine{}, po)
+		pm.AddDeadStoreEliminationPass()
-		if err != nil {
+		pm.AddAggressiveDCEPass()
-			t.Error("failed to run passes:", err)
+		pm.Run(mod)
-		}
 	})
 }

transform/optimizer.go

@@ -14,22 +14,54 @@ import (
 // OptimizePackage runs optimization passes over the LLVM module for the given
 // Go package.
 func OptimizePackage(mod llvm.Module, config *compileopts.Config) {
-	_, speedLevel, _ := config.OptLevel()
+	optLevel, sizeLevel, _ := config.OptLevels()
+
+	// Run function passes for each function in the module.
+	// These passes are intended to be run on each function right
+	// after they're created to reduce IR size (and maybe also for
+	// cache locality to improve performance), but for now they're
+	// run here for each function in turn. Maybe this can be
+	// improved in the future.
+	builder := llvm.NewPassManagerBuilder()
+	defer builder.Dispose()
+	builder.SetOptLevel(optLevel)
+	builder.SetSizeLevel(sizeLevel)
+	funcPasses := llvm.NewFunctionPassManagerForModule(mod)
+	defer funcPasses.Dispose()
+	builder.PopulateFunc(funcPasses)
+	funcPasses.InitializeFunc()
+	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
+		if fn.IsDeclaration() {
+			continue
+		}
+		funcPasses.RunFunc(fn)
+	}
+	funcPasses.FinalizeFunc()
 
 	// Run TinyGo-specific optimization passes.
-	if speedLevel > 0 {
+	if optLevel > 0 {
 		OptimizeMaps(mod)
 	}
 }
 
 // Optimize runs a number of optimization and transformation passes over the
 // given module. Some passes are specific to TinyGo, others are generic LLVM
-// passes.
+// passes. You can set a preferred performance (0-3) and size (0-2) level and
+// control the limits of the inliner (higher numbers mean more inlining, set it
+// to 0 to disable entirely).
 //
 // Please note that some optimizations are not optional, thus Optimize must
-// alwasy be run before emitting machine code.
+// alwasy be run before emitting machine code. Set all controls (optLevel,
-func Optimize(mod llvm.Module, config *compileopts.Config) []error {
+// sizeLevel, inlinerThreshold) to 0 to reduce the number of optimizations to a
-	optLevel, speedLevel, _ := config.OptLevel()
+// minimum.
+func Optimize(mod llvm.Module, config *compileopts.Config, optLevel, sizeLevel int, inlinerThreshold uint) []error {
+	builder := llvm.NewPassManagerBuilder()
+	defer builder.Dispose()
+	builder.SetOptLevel(optLevel)
+	builder.SetSizeLevel(sizeLevel)
+	if inlinerThreshold != 0 {
+		builder.UseInlinerWithThreshold(inlinerThreshold)
+	}
 
 	// Make sure these functions are kept in tact during TinyGo transformation passes.
 	for _, name := range functionsUsedInTransforms {
@@ -52,20 +84,23 @@ func Optimize(mod llvm.Module, config *compileopts.Config) []error {
 		}
 	}
 
-	if speedLevel > 0 {
+	if optLevel > 0 {
 		// Run some preparatory passes for the Go optimizer.
-		po := llvm.NewPassBuilderOptions()
+		goPasses := llvm.NewPassManager()
-		defer po.Dispose()
+		defer goPasses.Dispose()
-		err := mod.RunPasses("globaldce,globalopt,ipsccp,instcombine,adce,function-attrs", llvm.TargetMachine{}, po)
+		goPasses.AddGlobalDCEPass()
-		if err != nil {
+		goPasses.AddGlobalOptimizerPass()
-			return []error{fmt.Errorf("could not build pass pipeline: %w", err)}
+		goPasses.AddIPSCCPPass()
-		}
+		goPasses.AddInstructionCombiningPass() // necessary for OptimizeReflectImplements
+		goPasses.AddAggressiveDCEPass()
+		goPasses.AddFunctionAttrsPass()
+		goPasses.Run(mod)
 
 		// Run TinyGo-specific optimization passes.
 		OptimizeStringToBytes(mod)
 		OptimizeReflectImplements(mod)
 		OptimizeAllocs(mod, nil, nil)
-		err = LowerInterfaces(mod, config)
+		err := LowerInterfaces(mod, config)
 		if err != nil {
 			return []error{err}
 		}
@@ -78,10 +113,7 @@ func Optimize(mod llvm.Module, config *compileopts.Config) []error {
 		// After interfaces are lowered, there are many more opportunities for
 		// interprocedural optimizations. To get them to work, function
 		// attributes have to be updated first.
-		err = mod.RunPasses("globaldce,globalopt,ipsccp,instcombine,adce,function-attrs", llvm.TargetMachine{}, po)
+		goPasses.Run(mod)
-		if err != nil {
-			return []error{fmt.Errorf("could not build pass pipeline: %w", err)}
-		}
 
 		// Run TinyGo-specific interprocedural optimizations.
 		OptimizeAllocs(mod, config.Options.PrintAllocs, func(pos token.Position, msg string) {
@@ -102,12 +134,10 @@ func Optimize(mod llvm.Module, config *compileopts.Config) []error {
 		}
 
 		// Clean up some leftover symbols of the previous transformations.
-		po := llvm.NewPassBuilderOptions()
+		goPasses := llvm.NewPassManager()
-		defer po.Dispose()
+		defer goPasses.Dispose()
-		err = mod.RunPasses("globaldce", llvm.TargetMachine{}, po)
+		goPasses.AddGlobalDCEPass()
-		if err != nil {
+		goPasses.Run(mod)
-			return []error{fmt.Errorf("could not build pass pipeline: %w", err)}
-		}
 	}
 
 	if config.Scheduler() == "none" {
@@ -139,15 +169,23 @@ func Optimize(mod llvm.Module, config *compileopts.Config) []error {
 		fn.SetLinkage(llvm.InternalLinkage)
 	}
 
-	// Run the default pass pipeline.
+	// Run function passes again, because without it, llvm.coro.size.i32()
-	// TODO: set the PrepareForThinLTO flag somehow.
+	// doesn't get lowered.
-	po := llvm.NewPassBuilderOptions()
+	funcPasses := llvm.NewFunctionPassManagerForModule(mod)
-	defer po.Dispose()
+	defer funcPasses.Dispose()
-	passes := fmt.Sprintf("default<%s>", optLevel)
+	builder.PopulateFunc(funcPasses)
-	err := mod.RunPasses(passes, llvm.TargetMachine{}, po)
+	funcPasses.InitializeFunc()
-	if err != nil {
+	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		return []error{fmt.Errorf("could not build pass pipeline: %w", err)}
+		funcPasses.RunFunc(fn)
 	}
+	funcPasses.FinalizeFunc()
+
+	// Run module passes.
+	// TODO: somehow set the PrepareForThinLTO flag in the pass manager builder.
+	modPasses := llvm.NewPassManager()
+	defer modPasses.Dispose()
+	builder.Populate(modPasses)
+	modPasses.Run(mod)
 
 	hasGCPass := MakeGCStackSlots(mod)
 	if hasGCPass {

transform/rtcalls.go

@@ -28,44 +28,32 @@ func OptimizeStringToBytes(mod llvm.Module) {
 
 		// strptr is always constant because strings are always constant.
 
-		var pointerUses []llvm.Value
+		convertedAllUses := true
-		canConvertPointer := true
 		for _, use := range getUses(call) {
 			if use.IsAExtractValueInst().IsNil() {
 				// Expected an extractvalue, but this is something else.
-				canConvertPointer = false
+				convertedAllUses = false
 				continue
 			}
 			switch use.Type().TypeKind() {
 			case llvm.IntegerTypeKind:
 				// A length (len or cap). Propagate the length value.
-				// This can always be done because the byte slice is always the
-				// same length as the original string.
 				use.ReplaceAllUsesWith(strlen)
 				use.EraseFromParentAsInstruction()
 			case llvm.PointerTypeKind:
 				// The string pointer itself.
 				if !isReadOnly(use) {
-					// There is a store to the byte slice. This means that none
+					convertedAllUses = false
-					// of the pointer uses can't be propagated.
-					canConvertPointer = false
 					continue
 				}
-				// It may be that the pointer value can be propagated, if all of
+				use.ReplaceAllUsesWith(strptr)
-				// the pointer uses are readonly.
+				use.EraseFromParentAsInstruction()
-				pointerUses = append(pointerUses, use)
 			default:
 				// should not happen
 				panic("unknown return type of runtime.stringToBytes: " + use.Type().String())
 			}
 		}
-		if canConvertPointer {
+		if convertedAllUses {
-			// All pointer uses are readonly, so they can be converted.
-			for _, use := range pointerUses {
-				use.ReplaceAllUsesWith(strptr)
-				use.EraseFromParentAsInstruction()
-			}
-
 			// Call to runtime.stringToBytes can be eliminated: both the input
 			// and the output is constant.
 			call.EraseFromParentAsInstruction()

transform/testdata/allocs.out.ll

@@ -6,18 +6,18 @@ target triple = "armv7m-none-eabi"
 declare nonnull ptr @runtime.alloc(i32, ptr)
 
 define void @testInt() {
-  %stackalloc = alloca [4 x i8], align 4
+  %stackalloc.alloca = alloca [4 x i8], align 4
-  store [4 x i8] zeroinitializer, ptr %stackalloc, align 4
+  store [4 x i8] zeroinitializer, ptr %stackalloc.alloca, align 4
-  store i32 5, ptr %stackalloc, align 4
+  store i32 5, ptr %stackalloc.alloca, align 4
   ret void
 }
 
 define i16 @testArray() {
-  %stackalloc = alloca [6 x i8], align 2
+  %stackalloc.alloca = alloca [6 x i8], align 2
-  store [6 x i8] zeroinitializer, ptr %stackalloc, align 2
+  store [6 x i8] zeroinitializer, ptr %stackalloc.alloca, align 2
-  %alloc.1 = getelementptr i16, ptr %stackalloc, i32 1
+  %alloc.1 = getelementptr i16, ptr %stackalloc.alloca, i32 1
   store i16 5, ptr %alloc.1, align 2
-  %alloc.2 = getelementptr i16, ptr %stackalloc, i32 2
+  %alloc.2 = getelementptr i16, ptr %stackalloc.alloca, i32 2
   %val = load i16, ptr %alloc.2, align 2
   ret i16 %val
 }
@@ -35,9 +35,9 @@ define void @testEscapingCall2() {
 }
 
 define void @testNonEscapingCall() {
-  %stackalloc = alloca [4 x i8], align 4
+  %stackalloc.alloca = alloca [4 x i8], align 4
-  store [4 x i8] zeroinitializer, ptr %stackalloc, align 4
+  store [4 x i8] zeroinitializer, ptr %stackalloc.alloca, align 4
-  %val = call ptr @noescapeIntPtr(ptr %stackalloc)
+  %val = call ptr @noescapeIntPtr(ptr %stackalloc.alloca)
   ret void
 }
 
@@ -48,12 +48,12 @@ define ptr @testEscapingReturn() {
 
 define void @testNonEscapingLoop() {
 entry:
-  %stackalloc = alloca [4 x i8], align 4
+  %stackalloc.alloca = alloca [4 x i8], align 4
   br label %loop
 
 loop:                                             ; preds = %loop, %entry
-  store [4 x i8] zeroinitializer, ptr %stackalloc, align 4
+  store [4 x i8] zeroinitializer, ptr %stackalloc.alloca, align 4
-  %ptr = call ptr @noescapeIntPtr(ptr %stackalloc)
+  %ptr = call ptr @noescapeIntPtr(ptr %stackalloc.alloca)
   %result = icmp eq ptr null, %ptr
   br i1 %result, label %loop, label %end
 

transform/testdata/stringtobytes.ll

@@ -30,19 +30,3 @@ entry:
   call fastcc void @writeToSlice(ptr %1, i64 %2, i64 %3)
   ret void
 }
-
-; Test that pointer values are never propagated if there is even a single write
-; to the pointer value (but len/cap values still can be).
-define void @testReadSome() {
-entry:
-  %s = call fastcc { ptr, i64, i64 } @runtime.stringToBytes(ptr @str, i64 6)
-  %s.ptr = extractvalue { ptr, i64, i64 } %s, 0
-  %s.len = extractvalue { ptr, i64, i64 } %s, 1
-  %s.cap = extractvalue { ptr, i64, i64 } %s, 2
-  call fastcc void @writeToSlice(ptr %s.ptr, i64 %s.len, i64 %s.cap)
-  %s.ptr2 = extractvalue { ptr, i64, i64 } %s, 0
-  %s.len2 = extractvalue { ptr, i64, i64 } %s, 1
-  %s.cap2 = extractvalue { ptr, i64, i64 } %s, 2
-  call fastcc void @printSlice(ptr %s.ptr2, i64 %s.len2, i64 %s.cap2)
-  ret void
-}

transform/testdata/stringtobytes.out.ll

@@ -22,13 +22,3 @@ entry:
   call fastcc void @writeToSlice(ptr %1, i64 6, i64 6)
   ret void
 }
-
-define void @testReadSome() {
-entry:
-  %s = call fastcc { ptr, i64, i64 } @runtime.stringToBytes(ptr @str, i64 6)
-  %s.ptr = extractvalue { ptr, i64, i64 } %s, 0
-  call fastcc void @writeToSlice(ptr %s.ptr, i64 6, i64 6)
-  %s.ptr2 = extractvalue { ptr, i64, i64 } %s, 0
-  call fastcc void @printSlice(ptr %s.ptr2, i64 6, i64 6)
-  ret void
-}

transform/transform.go

@@ -22,7 +22,7 @@ import (
 // the -opt= compiler flag.
 func AddStandardAttributes(fn llvm.Value, config *compileopts.Config) {
 	ctx := fn.Type().Context()
-	_, _, sizeLevel := config.OptLevel()
+	_, sizeLevel, _ := config.OptLevels()
 	if sizeLevel >= 1 {
 		fn.AddFunctionAttr(ctx.CreateEnumAttribute(llvm.AttributeKindID("optsize"), 0))
 	}