clean up implementation

The compiler may generate calls to fminimum/fmaximum on some platforms.
Neither of the libm implementations we statically link against have these functions yet.
Implement them ourselves.

.github/workflows/build-macos.yml

@@ -16,14 +16,15 @@ jobs:
     name: build-macos
     strategy:
       matrix:
-        # macos-13: amd64 (oldest supported version as of 18-10-2024)
-        # macos-14: arm64 (oldest arm64 version)
-        os: [macos-13, macos-14]
+        # macos-14: arm64 (oldest supported version as of 18-11-2025)
+        # macos-15-intel: amd64 (last intel version to be supported by github runners)
+        # See https://github.com/actions/runner-images/issues/13046
+        os: [macos-14, macos-15-intel]
         include:
-          - os: macos-13
-            goarch: amd64
           - os: macos-14
             goarch: arm64
+          - os: macos-15-intel
+            goarch: amd64
     runs-on: ${{ matrix.os }}
     steps:
       - name: Install Dependencies
@@ -39,7 +40,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.1'
+          go-version: '1.25.7'
           cache: true
       - name: Restore LLVM source cache
         uses: actions/cache/restore@v4
@@ -134,7 +135,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.1'
+          go-version: '1.25.7'
           cache: true
       - name: Build TinyGo (LLVM ${{ matrix.version }})
         run: go install -tags=llvm${{ matrix.version }}

.github/workflows/docker.yml

@@ -31,7 +31,7 @@ jobs:
           sudo rm -rf /usr/local/share/boost
           df -h
       - name: Check out the repo
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           submodules: recursive
       - name: Set up Docker Buildx
@@ -58,7 +58,7 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
       - name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           context: .
           push: true
@@ -66,45 +66,3 @@ jobs:
           labels: ${{ steps.meta.outputs.labels }}
           cache-from: type=gha
           cache-to: type=gha,mode=max
-      - name: Trigger Drivers repo build on Github Actions
-        run: |
-          curl -X POST \
-          -H "Authorization: Bearer ${{secrets.GHA_ACCESS_TOKEN}}" \
-          -H "Accept: application/vnd.github.v3+json" \
-          https://api.github.com/repos/tinygo-org/drivers/actions/workflows/build.yml/dispatches \
-          -d '{"ref": "dev"}'
-      - name: Trigger Bluetooth repo build on Github Actions
-        run: |
-          curl -X POST \
-          -H "Authorization: Bearer ${{secrets.GHA_ACCESS_TOKEN}}" \
-          -H "Accept: application/vnd.github.v3+json" \
-          https://api.github.com/repos/tinygo-org/bluetooth/actions/workflows/linux.yml/dispatches \
-          -d '{"ref": "dev"}'
-      - name: Trigger TinyFS repo build on Github Actions
-        run: |
-          curl -X POST \
-          -H "Authorization: Bearer ${{secrets.GHA_ACCESS_TOKEN}}" \
-          -H "Accept: application/vnd.github.v3+json" \
-          https://api.github.com/repos/tinygo-org/tinyfs/actions/workflows/build.yml/dispatches \
-          -d '{"ref": "dev"}'
-      - name: Trigger TinyFont repo build on Github Actions
-        run: |
-          curl -X POST \
-          -H "Authorization: Bearer ${{secrets.GHA_ACCESS_TOKEN}}" \
-          -H "Accept: application/vnd.github.v3+json" \
-          https://api.github.com/repos/tinygo-org/tinyfont/actions/workflows/build.yml/dispatches \
-          -d '{"ref": "dev"}'
-      - name: Trigger TinyDraw repo build on Github Actions
-        run: |
-          curl -X POST \
-          -H "Authorization: Bearer ${{secrets.GHA_ACCESS_TOKEN}}" \
-          -H "Accept: application/vnd.github.v3+json" \
-          https://api.github.com/repos/tinygo-org/tinydraw/actions/workflows/build.yml/dispatches \
-          -d '{"ref": "dev"}'
-      - name: Trigger TinyTerm repo build on Github Actions
-        run: |
-          curl -X POST \
-          -H "Authorization: Bearer ${{secrets.GHA_ACCESS_TOKEN}}" \
-          -H "Accept: application/vnd.github.v3+json" \
-          https://api.github.com/repos/tinygo-org/tinyterm/actions/workflows/build.yml/dispatches \
-          -d '{"ref": "dev"}'

.github/workflows/linux.yml

@@ -31,7 +31,7 @@ jobs:
         # We're not on a multi-user machine, so this is safe.
         run: git config --global --add safe.directory "$GITHUB_WORKSPACE"
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           submodules: true
       - name: Extract TinyGo version
@@ -131,13 +131,13 @@ jobs:
     needs: build-linux
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           submodules: true
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Install wasmtime
         uses: bytecodealliance/actions/wasmtime/setup@v1
@@ -164,7 +164,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           submodules: true
       - name: Install apt dependencies
@@ -179,9 +179,9 @@ jobs:
               simavr \
               ninja-build
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Install Node.js
         uses: actions/setup-node@v4
@@ -284,7 +284,7 @@ jobs:
     needs: build-linux
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
       - name: Get TinyGo version
         id: version
         run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
@@ -296,9 +296,9 @@ jobs:
               g++-${{ matrix.toolchain }} \
               libc6-dev-${{ matrix.libc }}-cross
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Restore LLVM source cache
         uses: actions/cache/restore@v4

.github/workflows/llvm.yml

@@ -25,7 +25,7 @@ jobs:
       contents: read
     steps:
       - name: Check out the repo
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           submodules: recursive
       - name: Set up Docker Buildx
@@ -52,7 +52,7 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
       - name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           target: tinygo-llvm-build
           context: .

.github/workflows/nix.yml

@@ -21,7 +21,7 @@ jobs:
         # See: https://github.com/tinygo-org/tinygo/pull/4516#issuecomment-2416363668
         run: sudo apt-get remove llvm-18
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
       - name: Pull musl, bdwgc
         run: |
             git submodule update --init lib/musl lib/bdwgc
@@ -42,7 +42,7 @@ jobs:
           key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
           path: |
             llvm-project/compiler-rt
-      - uses: cachix/install-nix-action@v22
+      - uses: cachix/install-nix-action@v31
       - name: Test
         run: |
           nix develop --ignore-environment --keep HOME --command bash -c "go install && ~/go/bin/tinygo version && ~/go/bin/tinygo build -o test ./testdata/cgo"

.github/workflows/sizediff.yml

@@ -20,7 +20,7 @@ jobs:
         run: |
           echo "$HOME/go/bin" >> $GITHUB_PATH
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0 # fetch all history (no sparse checkout)
           submodules: true

.github/workflows/windows.yml

@@ -31,7 +31,7 @@ jobs:
         run: |
           scoop install ninja binaryen
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           submodules: true
       - name: Extract TinyGo version
@@ -39,9 +39,9 @@ jobs:
         shell: bash
         run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Restore cached LLVM source
         uses: actions/cache/restore@v4
@@ -143,11 +143,11 @@ jobs:
         run: |
           scoop install binaryen
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Download TinyGo build
         uses: actions/download-artifact@v4
@@ -173,11 +173,11 @@ jobs:
           maximum-size: 24GB
           disk-root: "C:"
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Download TinyGo build
         uses: actions/download-artifact@v4
@@ -209,11 +209,11 @@ jobs:
         run: |
           scoop install binaryen && scoop install wasmtime@29.0.1
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
       - name: Install Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.25.0'
+          go-version: '1.25.7'
           cache: true
       - name: Download TinyGo build
         uses: actions/download-artifact@v4

.gitmodules

@@ -16,7 +16,7 @@
 	url = https://github.com/WebAssembly/wasi-libc
 [submodule "lib/picolibc"]
 	path = lib/picolibc
-	url = https://github.com/keith-packard/picolibc.git
+	url = https://github.com/picolibc/picolibc.git
 [submodule "lib/stm32-svd"]
 	path = lib/stm32-svd
 	url = https://github.com/tinygo-org/stm32-svd

CHANGELOG.md

@@ -1,3 +1,104 @@
+0.40.1
+---
+* **machine**
+  - nrf: fix flash writes when SoftDevice is enabled
+* **runtime**
+  - runtime: avoid fixed math/rand sequence on RP2040/RP2350 (#5124)
+  - runtime: add calls to initRand() during run() for all schedulers
+  - runtime: call initRand() before initHeap() during initialization
+  - runtime: use rand_hwrng hardwareRand for RP2040/RP2350 (#5135)
+* **libs**
+  - picolibc: use updated location for git repo
+
+0.40.0
+---
+* **general**
+  - all: add full LLVM 20 support
+  - core: feat: enable //go:linkname pragma for globals
+  - core: feature: Add flag to ignore go compatibility matrix (#5078)
+  - chore: update version for 0.40 development cycle
+* **compiler**
+  - emit an error when the actual arch doesn't match GOARCH
+  - mark string parameters as readonly
+  - use Tarjan's SCC algorithm to detect loops for defer
+  - lower "large stack" limit to 16kb
+* **core**
+  - shrink bdwgc library
+  - Fix linker errors for runtime.vgetrandom and crypto/internal/sysrand.fatal
+  - fix: add TryLock to sync.RWMutex
+  - fix: correct linter issues exposed by the fix in #4679
+  - fix: don't hardcode success return state
+  - fix: expand RTT debugger compatibility
+  - internal/task (threads): save stack bounds instead of scanning under a lock
+  - internal/task: create detached threads and fix error handling
+  - interp: better errors when debugging interp
+  - transform (gc): create stack slots in callers of external functions
+  - internal/task: prevent semaphore resource leak for threads scheduler
+* **machine**
+  - cortexm: optimize code size for the HardFault_Handler
+  - fe310: add I2C pins for the HiFive1b
+  - clarify WriteAt semantics of BlockDevice
+  - fix deprecated AsmFull comment (#5005)
+  - make sure DMA buffers do not escape unnecessarily
+  - only enable USB-CDC when needed
+  - use larger SPI MAXCNT on nrf52833 and nrf52840
+  - fix: update m.queuedBytes when clamping output to avoid corrupting sentBytes
+  - fix: use int64 in ReadTemperature to avoid overflow
+  - fix(rp2): disable DBGPAUSE on startup
+  - fix(rp2): possible integer overflow while computing factors for SPI baudrate
+  - fix(rp2): reset spinlocks at startup
+  - fix(rp2): switch spinlock busy loop to wfe
+  - fix(rp2): use side-effect-free spinlocks
+  - nrf: add ADC_VDDH which is an ADC pin for VDDH
+  - nrf: don't block SPI transfer
+  - nrf: don't set PSELN, it's ignored in single ended mode anyway
+  - nrf: fix typo in ADC configuration
+  - nrf: refactor SoftDevice enabled check
+  - nrf: rename pwmPin to adcPin
+  - nrf: support flash operations while the SoftDevice is enabled
+  - rp2040: allow writing to the UART inside interrupts
+  - machine,nrf528: stop the bus only once on I2C bus error and ensures the first error is returned
+* **net**
+  - update submodule to latest commits
+* **runtime**
+  - (avr): fix infinite longjmp loop if stack is aligned to 256 bytes
+  - (gc_blocks.go): clear full size of allocation
+  - (gc_blocks.go): make sweep branchless
+  - (gc_blocks.go): simplify scanning logic
+  - (gc_blocks.go): use a linked stack to scan marked objects
+  - (gc_blocks.go): use best-fit allocation
+  - (gc_boehm.go): fix world already stopped check
+  - (wasm): scan the system stack
+  - fix sleep duration for long sleeps
+  - remove copied code for nrf52840
+  - src/syscall: update src buffer after write
+  - wasm: fix C realloc and optimize it a bit
+* **targets**
+  - add xiao-esp32s3 board target
+  - Add ESP32-S3 support (#5091)
+  - Added Gopher ARCADE board
+  - Create "pico2-ice" target board (#5062)
+* **build/test**
+  - Add testing.T.Context() and testing.B.Context()
+  - create separate go.mod file for testing dependencies for wasm tests that use Chromium headless browser to avoid use of older incompatible version.
+  - go back to normal scheduler instead of tasks scheduler for macos CI
+  - update CI to use Go 1.25.5
+  - update macOS GH actions builds to handle sunset of macOS 13
+  - use task scheduler on macOS builds to avoid test race condition lockups
+  - update all CI builds to use latest stable Go release. Also update some of the actions to their latest releases.
+  - update GH actions builds to use Go 1.25.4
+  - uninstall cmake before install
+  - fix: point the submodule for musl-lib to a mirror in the TinyGo GitHub org
+  - fix: remove macOS 15 from CI build matrix (conflicts with macOS 14 build)
+  - fix: separate host expected bytes from device intended bytes
+  - fix/typo: makeESPFirmwareImage
+  - make: GNUmakefile: shrink TinyGo binaries on Linux
+  - move the directory list into a variable
+  - several improvements to the macOS GH actions build
+  - Fix for #4678: top-level 'make lint' wasn't working
+  - fix: increase the timeout for chromedp to connect to the headless browser used for running the wasm tests.
+  - testdata: some more packages for the test corpus
+
 0.39.0
 ---
 * **general**

GNUmakefile

@@ -481,6 +481,7 @@ TEST_IOFS := false
 endif
 
 TEST_SKIP_FLAG := -skip='TestExtraMethods|TestParseAndBytesRoundTrip/P256/Generic'
+TEST_ADDITIONAL_FLAGS ?=
 
 # Test known-working standard library packages.
 # TODO: parallelize, and only show failing tests (no implied -v flag).
@@ -488,7 +489,7 @@ TEST_SKIP_FLAG := -skip='TestExtraMethods|TestParseAndBytesRoundTrip/P256/Generi
 tinygo-test:
 	@# TestExtraMethods: used by many crypto packages and uses reflect.Type.Method which is not implemented.
 	@# TestParseAndBytesRoundTrip/P256/Generic: relies on t.Skip() which is not implemented
-	$(TINYGO) test $(TEST_SKIP_FLAG) $(TEST_PACKAGES_HOST) $(TEST_PACKAGES_SLOW)
+	$(TINYGO) test $(TEST_ADDITIONAL_FLAGS) $(TEST_SKIP_FLAG) $(TEST_PACKAGES_HOST) $(TEST_PACKAGES_SLOW)
 	@# io/fs requires os.ReadDir, not yet supported on windows or wasi. It also
 	@# requires a large stack-size. Hence, io/fs is only run conditionally.
 	@# For more details, see the comments on issue #3143.
@@ -621,6 +622,8 @@ smoketest: testchdir
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-rp2040      examples/device-id
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico2-ice           examples/blinky1
+	@$(MD5SUM) test.hex
 	# test simulated boards on play.tinygo.org
 ifneq ($(WASM), 0)
 	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=arduino              examples/blinky1
@@ -811,6 +814,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=waveshare-rp2040-tiny examples/echo
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=vicharak_shrike-lite examples/echo
+	@$(MD5SUM) test.hex
 	# test pwm
 	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/pwm
 	@$(MD5SUM) test.hex
@@ -893,6 +898,10 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=digispark           examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=digispark           examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=digispark           examples/mcp3008
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=digispark -gc=leaking examples/blinky1
 	@$(MD5SUM) test.hex
 ifneq ($(XTENSA), 0)
@@ -912,6 +921,12 @@ ifneq ($(XTENSA), 0)
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target mch2022             examples/machinetest
 	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=xiao-esp32s3   		examples/blinky1
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=xiao-esp32s3   		examples/mcp3008
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=esp32s3-wroom1	    examples/mcp3008
+	@$(MD5SUM) test.bin
 endif
 	$(TINYGO) build -size short -o test.bin -target=esp-c3-32s-kit      examples/blinky1
 	@$(MD5SUM) test.bin

LICENSE

@@ -1,7 +1,8 @@
-Copyright (c) 2018-2025 The TinyGo Authors. All rights reserved.
+Copyright (c) 2018-2026 The TinyGo Authors. All rights reserved.
 
 TinyGo includes portions of the Go standard library.
-Copyright (c) 2009-2024 The Go Authors. All rights reserved.
+Copyright 2009 The Go Authors. All rights reserved.
+See https://github.com/golang/go/blob/master/LICENSE for license information.
 
 TinyGo includes portions of LLVM, which is under the Apache License v2.0 with
 LLVM Exceptions. See https://llvm.org/LICENSE.txt for license information.

README.md

@@ -6,6 +6,9 @@ TinyGo is a Go compiler intended for use in small places such as microcontroller
 
 It reuses libraries used by the [Go language tools](https://golang.org/pkg/go/) alongside [LLVM](http://llvm.org) to provide an alternative way to compile programs written in the Go programming language.
 
+> [!IMPORTANT]
+> You can help TinyGo with a financial contribution using OpenCollective. Please see https://opencollective.com/tinygo for more information. Thank you!
+
 ## Embedded
 
 Here is an example program that blinks the built-in LED when run directly on any supported board with onboard LED:
@@ -63,7 +66,7 @@ tinygo build -buildmode=c-shared -o add.wasm -target=wasip1 add.go
 You can also use the same syntax as Go 1.24+:
 
 ```shell
-GOARCH=wasip1 GOOS=wasm tinygo build -buildmode=c-shared -o add.wasm add.go
+GOOS=wasip1 GOARCH=wasm tinygo build -buildmode=c-shared -o add.wasm add.go
 ```
 
 ## Installation

builder/bdwgc.go

@@ -30,7 +30,10 @@ var BoehmGC = Library{
 			// Use a minimal environment.
 			"-DNO_MSGBOX_ON_ERROR", // don't call MessageBoxA on Windows
 			"-DDONT_USE_ATEXIT",
-			"-DNO_GETENV",
+			"-DNO_GETENV",          // smaller binary, more predictable configuration
+			"-DNO_CLOCK",           // don't use system clock
+			"-DNO_DEBUGGING",       // reduce code size
+			"-DGC_NO_FINALIZATION", // finalization is not used at the moment
 
 			// Special flag to work around the lack of __data_start in ld.lld.
 			// TODO: try to fix this in LLVM/lld directly so we don't have to
@@ -39,6 +42,8 @@ var BoehmGC = Library{
 
 			// Do not scan the stack. We have our own mechanism to do this.
 			"-DSTACK_NOT_SCANNED",
+			"-DNO_PROC_STAT",  // we scan the stack manually (don't read /proc/self/stat on Linux)
+			"-DSTACKBOTTOM=0", // dummy value, we scan the stack manually
 
 			// Assertions can be enabled while debugging GC issues.
 			//"-DGC_ASSERTIONS",
@@ -63,7 +68,6 @@ var BoehmGC = Library{
 			"blacklst.c",
 			"dbg_mlc.c",
 			"dyn_load.c",
-			"finalize.c",
 			"headers.c",
 			"mach_dep.c",
 			"malloc.c",
@@ -71,7 +75,6 @@ var BoehmGC = Library{
 			"mark_rts.c",
 			"misc.c",
 			"new_hblk.c",
-			"obj_map.c",
 			"os_dep.c",
 			"reclaim.c",
 		}

builder/build.go

@@ -19,6 +19,7 @@ import (
 	"os/exec"
 	"path/filepath"
 	"runtime"
+	"slices"
 	"sort"
 	"strconv"
 	"strings"
@@ -281,9 +282,13 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 				allFiles[file.Name] = append(allFiles[file.Name], file)
 			}
 		}
-		for name, files := range allFiles {
-			name := name
-			files := files
+		// Sort embedded files by name to maintain output determinism.
+		embedNames := make([]string, 0, len(allFiles))
+		for _, files := range allFiles {
+			embedNames = append(embedNames, files[0].Name)
+		}
+		slices.Sort(embedNames)
+		for _, name := range embedNames {
 			job := &compileJob{
 				description: "make object file for " + name,
 				run: func(job *compileJob) error {
@@ -298,7 +303,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					sum := sha256.Sum256(data)
 					hexSum := hex.EncodeToString(sum[:16])
 
-					for _, file := range files {
+					for _, file := range allFiles[name] {
 						file.Size = uint64(len(data))
 						file.Hash = hexSum
 						if file.NeedsData {
@@ -1042,7 +1047,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		if err != nil {
 			return result, err
 		}
-	case "esp32", "esp32-img", "esp32c3", "esp8266":
+	case "esp32", "esp32-img", "esp32c3", "esp32s3", "esp8266":
 		// Special format for the ESP family of chips (parsed by the ROM
 		// bootloader).
 		result.Binary = filepath.Join(tmpdir, "main"+outext)

builder/builder_test.go

@@ -28,6 +28,7 @@ func TestClangAttributes(t *testing.T) {
 		"cortex-m4",
 		"cortex-m7",
 		"esp32c3",
+		"esp32s3",
 		"fe310",
 		"gameboy-advance",
 		"k210",

builder/config.go

@@ -33,10 +33,13 @@ func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
 	if err != nil {
 		return nil, err
 	}
-	if gorootMajor != 1 || gorootMinor < minorMin || gorootMinor > minorMax {
-		// Note: when this gets updated, also update the Go compatibility matrix:
-		// https://github.com/tinygo-org/tinygo-site/blob/dev/content/docs/reference/go-compat-matrix.md
-		return nil, fmt.Errorf("requires go version 1.%d through 1.%d, got go%d.%d", minorMin, minorMax, gorootMajor, gorootMinor)
+
+	if options.GoCompatibility {
+		if gorootMajor != 1 || gorootMinor < minorMin || gorootMinor > minorMax {
+			// Note: when this gets updated, also update the Go compatibility matrix:
+			// https://github.com/tinygo-org/tinygo-site/blob/dev/content/docs/reference/go-compat-matrix.md
+			return nil, fmt.Errorf("requires go version 1.%d through 1.%d, got go%d.%d", minorMin, minorMax, gorootMajor, gorootMinor)
+		}
 	}
 
 	// Check that the Go toolchain version isn't too new, if we haven't been

builder/esp.go

@@ -100,11 +100,12 @@ func makeESPFirmwareImage(infile, outfile, format string) error {
 	chip_id := map[string]uint16{
 		"esp32":   0x0000,
 		"esp32c3": 0x0005,
+		"esp32s3": 0x0009,
 	}[chip]
 
 	// Image header.
 	switch chip {
-	case "esp32", "esp32c3":
+	case "esp32", "esp32c3", "esp32s3":
 		// Header format:
 		// https://github.com/espressif/esp-idf/blob/v4.3/components/bootloader_support/include/esp_app_format.h#L71
 		// Note: not adding a SHA256 hash as the binary is modified by

builder/sizes_test.go

@@ -42,9 +42,9 @@ 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", 3884, 280, 0, 2268},
-		{"microbit", "examples/serial", 2924, 388, 8, 2272},
-		{"wioterminal", "examples/pininterrupt", 7365, 1491, 116, 6912},
+		{"hifive1b", "examples/echo", 3668, 280, 0, 2244},
+		{"microbit", "examples/serial", 2694, 342, 8, 2248},
+		{"wioterminal", "examples/pininterrupt", 6837, 1491, 120, 6888},
 
 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the

compileopts/config.go

@@ -226,7 +226,7 @@ func (c *Config) StackSize() uint64 {
 
 // MaxStackAlloc returns the size of the maximum allocation to put on the stack vs heap.
 func (c *Config) MaxStackAlloc() uint64 {
-	if c.StackSize() > 32*1024 {
+	if c.StackSize() >= 16*1024 {
 		return 1024
 	}
 

compileopts/options.go

@@ -59,6 +59,7 @@ type Options struct {
 	WITPackage      string // pass through to wasm-tools component embed invocation
 	WITWorld        string // pass through to wasm-tools component embed -w option
 	ExtLDFlags      []string
+	GoCompatibility bool // enable to check for Go version compatibility
 }
 
 // Verify performs a validation on the given options, raising an error if options are not valid.

compiler/asserts.go

@@ -245,7 +245,7 @@ func (b *builder) createRuntimeAssert(assert llvm.Value, blockPrefix, assertFunc
 	// current insert position.
 	faultBlock := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".throw")
 	nextBlock := b.insertBasicBlock(blockPrefix + ".next")
-	b.blockExits[b.currentBlock] = nextBlock // adjust outgoing block for phi nodes
+	b.currentBlockInfo.exit = nextBlock // adjust outgoing block for phi nodes
 
 	// Now branch to the out-of-bounds or the regular block.
 	b.CreateCondBr(assert, faultBlock, nextBlock)

compiler/calls.go

@@ -32,6 +32,9 @@ const (
 	// Whether this is a full or partial Go parameter (int, slice, etc).
 	// The extra context parameter is not a Go parameter.
 	paramIsGoParam = 1 << iota
+
+	// Whether this is a readonly parameter (for example, a string pointer).
+	paramIsReadonly
 )
 
 // createRuntimeCallCommon creates a runtime call. Use createRuntimeCall or
@@ -167,6 +170,7 @@ func (c *compilerContext) flattenAggregateType(t llvm.Type, name string, goType
 				continue
 			}
 			suffix := strconv.Itoa(i)
+			isString := false
 			if goType != nil {
 				// Try to come up with a good suffix for this struct field,
 				// depending on which Go type it's based on.
@@ -183,12 +187,16 @@ func (c *compilerContext) flattenAggregateType(t llvm.Type, name string, goType
 						suffix = []string{"r", "i"}[i]
 					case types.String:
 						suffix = []string{"data", "len"}[i]
+						isString = true
 					}
 				case *types.Signature:
 					suffix = []string{"context", "funcptr"}[i]
 				}
 			}
 			subInfos := c.flattenAggregateType(subfield, name+"."+suffix, extractSubfield(goType, i))
+			if isString {
+				subInfos[0].flags |= paramIsReadonly
+			}
 			paramInfos = append(paramInfos, subInfos...)
 		}
 		return paramInfos

compiler/compiler.go

@@ -152,10 +152,12 @@ type builder struct {
 	llvmFnType        llvm.Type
 	llvmFn            llvm.Value
 	info              functionInfo
-	locals            map[ssa.Value]llvm.Value            // local variables
-	blockEntries      map[*ssa.BasicBlock]llvm.BasicBlock // a *ssa.BasicBlock may be split up
-	blockExits        map[*ssa.BasicBlock]llvm.BasicBlock // these are the exit blocks
+	locals            map[ssa.Value]llvm.Value // local variables
+	blockInfo         []blockInfo
 	currentBlock      *ssa.BasicBlock
+	currentBlockInfo  *blockInfo
+	tarjanStack       []uint
+	tarjanIndex       uint
 	phis              []phiNode
 	deferPtr          llvm.Value
 	deferFrame        llvm.Value
@@ -187,11 +189,22 @@ func newBuilder(c *compilerContext, irbuilder llvm.Builder, f *ssa.Function) *bu
 		info:            c.getFunctionInfo(f),
 		locals:          make(map[ssa.Value]llvm.Value),
 		dilocals:        make(map[*types.Var]llvm.Metadata),
-		blockEntries:    make(map[*ssa.BasicBlock]llvm.BasicBlock),
-		blockExits:      make(map[*ssa.BasicBlock]llvm.BasicBlock),
 	}
 }
 
+type blockInfo struct {
+	// entry is the LLVM basic block corresponding to the start of this *ssa.Block.
+	entry llvm.BasicBlock
+
+	// exit is the LLVM basic block corresponding to the end of this *ssa.Block.
+	// It will be different than entry if any of the block's instructions contain internal branches.
+	exit llvm.BasicBlock
+
+	// tarjan holds state for applying Tarjan's strongly connected components algorithm to the CFG.
+	// This is used by defer.go to determine whether to stack- or heap-allocate defer data.
+	tarjan tarjanNode
+}
+
 type deferBuiltin struct {
 	callName string
 	pos      token.Pos
@@ -1220,14 +1233,29 @@ func (b *builder) createFunctionStart(intrinsic bool) {
 		// intrinsic (like an atomic operation). Create the entry block
 		// manually.
 		entryBlock = b.ctx.AddBasicBlock(b.llvmFn, "entry")
-	} else {
-		for _, block := range b.fn.DomPreorder() {
-			llvmBlock := b.ctx.AddBasicBlock(b.llvmFn, block.Comment)
-			b.blockEntries[block] = llvmBlock
-			b.blockExits[block] = llvmBlock
+		// Intrinsics may create internal branches (e.g. nil checks).
+		// They will attempt to access b.currentBlockInfo to update the exit block.
+		// Create some fake block info for them to access.
+		blockInfo := []blockInfo{
+			{
+				entry: entryBlock,
+				exit:  entryBlock,
+			},
 		}
+		b.blockInfo = blockInfo
+		b.currentBlockInfo = &blockInfo[0]
+	} else {
+		blocks := b.fn.Blocks
+		blockInfo := make([]blockInfo, len(blocks))
+		for _, block := range b.fn.DomPreorder() {
+			info := &blockInfo[block.Index]
+			llvmBlock := b.ctx.AddBasicBlock(b.llvmFn, block.Comment)
+			info.entry = llvmBlock
+			info.exit = llvmBlock
+		}
+		b.blockInfo = blockInfo
 		// Normal functions have an entry block.
-		entryBlock = b.blockEntries[b.fn.Blocks[0]]
+		entryBlock = blockInfo[0].entry
 	}
 	b.SetInsertPointAtEnd(entryBlock)
 
@@ -1323,8 +1351,9 @@ func (b *builder) createFunction() {
 		if b.DumpSSA {
 			fmt.Printf("%d: %s:\n", block.Index, block.Comment)
 		}
-		b.SetInsertPointAtEnd(b.blockEntries[block])
 		b.currentBlock = block
+		b.currentBlockInfo = &b.blockInfo[block.Index]
+		b.SetInsertPointAtEnd(b.currentBlockInfo.entry)
 		for _, instr := range block.Instrs {
 			if instr, ok := instr.(*ssa.DebugRef); ok {
 				if !b.Debug {
@@ -1384,7 +1413,7 @@ func (b *builder) createFunction() {
 		block := phi.ssa.Block()
 		for i, edge := range phi.ssa.Edges {
 			llvmVal := b.getValue(edge, getPos(phi.ssa))
-			llvmBlock := b.blockExits[block.Preds[i]]
+			llvmBlock := b.blockInfo[block.Preds[i].Index].exit
 			phi.llvm.AddIncoming([]llvm.Value{llvmVal}, []llvm.BasicBlock{llvmBlock})
 		}
 	}
@@ -1498,11 +1527,11 @@ func (b *builder) createInstruction(instr ssa.Instruction) {
 	case *ssa.If:
 		cond := b.getValue(instr.Cond, getPos(instr))
 		block := instr.Block()
-		blockThen := b.blockEntries[block.Succs[0]]
-		blockElse := b.blockEntries[block.Succs[1]]
+		blockThen := b.blockInfo[block.Succs[0].Index].entry
+		blockElse := b.blockInfo[block.Succs[1].Index].entry
 		b.CreateCondBr(cond, blockThen, blockElse)
 	case *ssa.Jump:
-		blockJump := b.blockEntries[instr.Block().Succs[0]]
+		blockJump := b.blockInfo[instr.Block().Succs[0].Index].entry
 		b.CreateBr(blockJump)
 	case *ssa.MapUpdate:
 		m := b.getValue(instr.Map, getPos(instr))
@@ -1570,7 +1599,8 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		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")
+		elemLayout := b.createObjectLayout(elemType, pos)
+		result := b.createRuntimeCall("sliceAppend", []llvm.Value{srcBuf, elemsBuf, srcLen, srcCap, elemsLen, elemSize, elemLayout}, "append.new")
 		newPtr := b.CreateExtractValue(result, 0, "append.newPtr")
 		newLen := b.CreateExtractValue(result, 1, "append.newLen")
 		newCap := b.CreateExtractValue(result, 2, "append.newCap")
@@ -1652,13 +1682,41 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 	case "copy":
 		dst := argValues[0]
 		src := argValues[1]
+		// Fetch the lengths.
 		dstLen := b.CreateExtractValue(dst, 1, "copy.dstLen")
 		srcLen := b.CreateExtractValue(src, 1, "copy.srcLen")
-		dstBuf := b.CreateExtractValue(dst, 0, "copy.dstArray")
-		srcBuf := b.CreateExtractValue(src, 0, "copy.srcArray")
+		// Find the minimum of the lengths.
+		minFuncName := "llvm.umin.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+		minFunc := b.mod.NamedFunction(minFuncName)
+		if minFunc.IsNil() {
+			fnType := llvm.FunctionType(b.uintptrType, []llvm.Type{b.uintptrType, b.uintptrType}, false)
+			minFunc = llvm.AddFunction(b.mod, minFuncName, fnType)
+		}
+		minLen := b.CreateCall(minFunc.GlobalValueType(), minFunc, []llvm.Value{dstLen, srcLen}, "copy.n")
+		// Multiply the length by the element size.
 		elemType := b.getLLVMType(argTypes[0].Underlying().(*types.Slice).Elem())
 		elemSize := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(elemType), false)
-		return b.createRuntimeCall("sliceCopy", []llvm.Value{dstBuf, srcBuf, dstLen, srcLen, elemSize}, "copy.n"), nil
+		// NOTE: This is also NSW when uintptr is int, but we can only choose one through the C API?
+		size := b.CreateNUWMul(minLen, elemSize, "copy.size")
+		// Fetch the pointers.
+		dstBuf := b.CreateExtractValue(dst, 0, "copy.dstPtr")
+		srcBuf := b.CreateExtractValue(src, 0, "copy.srcPtr")
+		// Create a memcpy.
+		call := b.createMemCopy("memmove", dstBuf, srcBuf, size)
+		align := b.targetData.ABITypeAlignment(elemType)
+		if align > 1 {
+			// Apply the type's alignment to the arguments.
+			// LLVM sometimes turns constant-length moves into loads and stores.
+			// It may use this alignment for the created loads and stores.
+			alignAttr := b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align))
+			call.AddCallSiteAttribute(1, alignAttr)
+			call.AddCallSiteAttribute(2, alignAttr)
+		}
+		// Extend and return the copied length.
+		if b.targetData.TypeAllocSize(minLen.Type()) < b.targetData.TypeAllocSize(b.intType) {
+			minLen = b.CreateZExt(minLen, b.intType, "copy.n.zext")
+		}
+		return minLen, nil
 	case "delete":
 		m := argValues[0]
 		key := argValues[1]
@@ -1686,20 +1744,66 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		return llvmLen, nil
 	case "min", "max":
 		// min and max builtins, added in Go 1.21.
-		// We can simply reuse the existing binop comparison code, which has all
-		// the edge cases figured out already.
-		tok := token.LSS
-		if callName == "max" {
-			tok = token.GTR
-		}
-		result := argValues[0]
-		typ := argTypes[0]
-		for _, arg := range argValues[1:] {
-			cmp, err := b.createBinOp(tok, typ, typ, result, arg, pos)
-			if err != nil {
-				return result, err
+		// Find the corresponding intrinsic name.
+		ty := argTypes[0].Underlying().(*types.Basic)
+		llvmType := b.getLLVMType(ty)
+		info := ty.Info()
+		var prefix, delimeter, typeName string
+		if info&types.IsInteger != 0 {
+			// This is an integer value.
+			// Use the LLVM int min/max intrinsics.
+			prefix = "llvm.s"
+			if info&types.IsUnsigned != 0 {
+				prefix = "llvm.u"
 			}
-			result = b.CreateSelect(cmp, result, arg, "")
+			delimeter = ".i"
+			typeName = strconv.Itoa(llvmType.IntTypeWidth())
+		} else {
+			switch ty.Kind() {
+			case types.String:
+				// Strings do not have an equivalent intrinsic.
+				// Implement with compares and selects.
+				tok := token.LSS
+				if callName == "max" {
+					tok = token.GTR
+				}
+				result := argValues[0]
+				typ := argTypes[0]
+				for _, arg := range argValues[1:] {
+					cmp, err := b.createBinOp(tok, typ, typ, result, arg, pos)
+					if err != nil {
+						return result, err
+					}
+					result = b.CreateSelect(cmp, result, arg, "")
+				}
+				return result, nil
+			case types.Float32:
+				typeName = "f32"
+			case types.Float64:
+				typeName = "f64"
+			default:
+				return llvm.Value{}, b.makeError(pos, "todo: min/max: unknown type")
+			}
+			// There are a few edge cases with floating point min/max:
+			// min(-0.0, +0.0) = -0.0
+			// min(NaN, number) = NaN
+			// The llvm.minimum.*/llvm.maximum.* intrinsics match this behavior.
+			// Neither Go nor LLVM defines the bit representation of resulting NaNs.
+			prefix = "llvm."
+			delimeter = "imum."
+		}
+		intrinsicName := prefix + callName + delimeter + typeName
+		// Find or create the intrinsic.
+		llvmFn := b.mod.NamedFunction(intrinsicName)
+		if llvmFn.IsNil() {
+			fnType := llvm.FunctionType(llvmType, []llvm.Type{llvmType, llvmType}, false)
+			llvmFn = llvm.AddFunction(b.mod, intrinsicName, fnType)
+		}
+		// Call the intrinsic repeatedly to merge the arguments.
+		callType := llvmFn.GlobalValueType()
+		result := argValues[0]
+		for _, arg := range argValues[1:] {
+			result = b.CreateCall(callType, llvmFn, []llvm.Value{result, arg}, "")
 		}
 		return result, nil
 	case "panic":

compiler/defer.go

@@ -100,7 +100,7 @@ func (b *builder) createLandingPad() {
 
 	// Continue at the 'recover' block, which returns to the parent in an
 	// appropriate way.
-	b.CreateBr(b.blockEntries[b.fn.Recover])
+	b.CreateBr(b.blockInfo[b.fn.Recover.Index].entry)
 }
 
 // Create a checkpoint (similar to setjmp). This emits inline assembly that
@@ -234,41 +234,108 @@ func (b *builder) createInvokeCheckpoint() {
 	continueBB := b.insertBasicBlock("")
 	b.CreateCondBr(isZero, continueBB, b.landingpad)
 	b.SetInsertPointAtEnd(continueBB)
-	b.blockExits[b.currentBlock] = continueBB
+	b.currentBlockInfo.exit = continueBB
 }
 
-// isInLoop checks if there is a path from a basic block to itself.
-func isInLoop(start *ssa.BasicBlock) bool {
-	// Use a breadth-first search to scan backwards through the block graph.
-	queue := []*ssa.BasicBlock{start}
-	checked := map[*ssa.BasicBlock]struct{}{}
+// isInLoop checks if there is a path from the current block to itself.
+// Use Tarjan's strongly connected components algorithm to search for cycles.
+// A one-node SCC is a cycle iff there is an edge from the node to itself.
+// A multi-node SCC is always a cycle.
+// https://en.wikipedia.org/wiki/Tarjan%27s_strongly_connected_components_algorithm
+func (b *builder) isInLoop() bool {
+	if b.currentBlockInfo.tarjan.lowLink == 0 {
+		b.strongConnect(b.currentBlock)
+	}
+	return b.currentBlockInfo.tarjan.cyclic
+}
 
-	for len(queue) > 0 {
-		// pop a block off of the queue
-		block := queue[len(queue)-1]
-		queue = queue[:len(queue)-1]
+func (b *builder) strongConnect(block *ssa.BasicBlock) {
+	// Assign a new index.
+	// Indices start from 1 so that 0 can be used as a sentinel.
+	assignedIndex := b.tarjanIndex + 1
+	b.tarjanIndex = assignedIndex
 
-		// Search through predecessors.
-		// Searching backwards means that this is pretty fast when the block is close to the start of the function.
-		// Defers are often placed near the start of the function.
-		for _, pred := range block.Preds {
-			if pred == start {
-				// cycle found
-				return true
-			}
+	// Apply the new index.
+	blockIndex := block.Index
+	node := &b.blockInfo[blockIndex].tarjan
+	node.lowLink = assignedIndex
 
-			if _, ok := checked[pred]; ok {
-				// block already checked
-				continue
-			}
+	// Push the node onto the stack.
+	node.onStack = true
+	b.tarjanStack = append(b.tarjanStack, uint(blockIndex))
 
-			// add to queue and checked map
-			queue = append(queue, pred)
-			checked[pred] = struct{}{}
+	// Process the successors.
+	for _, successor := range block.Succs {
+		// Look up the successor's state.
+		successorIndex := successor.Index
+		if successorIndex == blockIndex {
+			// Handle a self-cycle specially.
+			node.cyclic = true
+			continue
+		}
+		successorNode := &b.blockInfo[successorIndex].tarjan
+
+		switch {
+		case successorNode.lowLink == 0:
+			// This node has not yet been visisted.
+			b.strongConnect(successor)
+
+		case !successorNode.onStack:
+			// This node has been visited, but is in a different SCC.
+			// Ignore it, and do not update lowLink.
+			continue
+		}
+
+		// Update the lowLink index.
+		// This always uses the min-of-lowlink instead of using index in the on-stack case.
+		// This is done for two reasons:
+		// 1. The lowLink update can be shared between the new-node and on-stack cases.
+		// 2. The assigned index does not need to be saved - it is only needed for root node detection.
+		if successorNode.lowLink < node.lowLink {
+			node.lowLink = successorNode.lowLink
 		}
 	}
 
-	return false
+	if node.lowLink == assignedIndex {
+		// This is a root node.
+		// Pop the SCC off the stack.
+		stack := b.tarjanStack
+		top := stack[len(stack)-1]
+		stack = stack[:len(stack)-1]
+		blocks := b.blockInfo
+		topNode := &blocks[top].tarjan
+		topNode.onStack = false
+
+		if top != uint(blockIndex) {
+			// The root node is not the only node in the SCC.
+			// Mark all nodes in this SCC as cyclic.
+			topNode.cyclic = true
+			for top != uint(blockIndex) {
+				top = stack[len(stack)-1]
+				stack = stack[:len(stack)-1]
+				topNode = &blocks[top].tarjan
+				topNode.onStack = false
+				topNode.cyclic = true
+			}
+		}
+
+		b.tarjanStack = stack
+	}
+}
+
+// tarjanNode holds per-block state for isInLoop and strongConnect.
+type tarjanNode struct {
+	// lowLink is the index of the first visited node that is reachable from this block.
+	// The lowLink indices are assigned by the SCC search, and do not correspond to b.Index.
+	// A lowLink of 0 is used as a sentinel to mark a node which has not yet been visited.
+	lowLink uint
+
+	// onStack tracks whether this node is currently on the SCC search stack.
+	onStack bool
+
+	// cyclic indicates whether this block is in a loop.
+	// If lowLink is 0, strongConnect must be called before reading this field.
+	cyclic bool
 }
 
 // createDefer emits a single defer instruction, to be run when this function
@@ -410,7 +477,10 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 
 	// Put this struct in an allocation.
 	var alloca llvm.Value
-	if !isInLoop(instr.Block()) {
+	if instr.Block() != b.currentBlock {
+		panic("block mismatch")
+	}
+	if !b.isInLoop() {
 		// This can safely use a stack allocation.
 		alloca = llvmutil.CreateEntryBlockAlloca(b.Builder, deferredCallType, "defer.alloca")
 	} else {

compiler/gc.go

@@ -99,6 +99,9 @@ func typeHasPointers(t llvm.Type) bool {
 		}
 		return false
 	case llvm.ArrayTypeKind:
+		if t.ArrayLength() == 0 {
+			return false
+		}
 		if typeHasPointers(t.ElementType()) {
 			return true
 		}

compiler/inlineasm.go

@@ -38,10 +38,10 @@ func (b *builder) createInlineAsm(args []ssa.Value) (llvm.Value, error) {
 // provided immediately. For example:
 //
 //	arm.AsmFull(
-//	    "str {value}, {result}",
+//	    "str {value}, [{result}]",
 //	    map[string]interface{}{
-//	        "value":  1
-//	        "result": &dest,
+//	        "value":  1,
+//	        "result": uintptr(unsafe.Pointer(&dest)),
 //	    })
 func (b *builder) createInlineAsmFull(instr *ssa.CallCommon) (llvm.Value, error) {
 	asmString := constant.StringVal(instr.Args[0].(*ssa.Const).Value)

compiler/interface.go

@@ -737,7 +737,7 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
 	prevBlock := b.GetInsertBlock()
 	okBlock := b.insertBasicBlock("typeassert.ok")
 	nextBlock := b.insertBasicBlock("typeassert.next")
-	b.blockExits[b.currentBlock] = nextBlock // adjust outgoing block for phi nodes
+	b.currentBlockInfo.exit = nextBlock // adjust outgoing block for phi nodes
 	b.CreateCondBr(commaOk, okBlock, nextBlock)
 
 	// Retrieve the value from the interface if the type assert was

compiler/intrinsics.go

@@ -27,6 +27,8 @@ func (b *builder) defineIntrinsicFunction() {
 		b.createStackSaveImpl()
 	case name == "runtime.KeepAlive":
 		b.createKeepAliveImpl()
+	case name == "machine.keepAliveNoEscape":
+		b.createMachineKeepAliveImpl()
 	case strings.HasPrefix(name, "runtime/volatile.Load"):
 		b.createVolatileLoad()
 	case strings.HasPrefix(name, "runtime/volatile.Store"):
@@ -48,19 +50,24 @@ func (b *builder) defineIntrinsicFunction() {
 // and will otherwise be lowered to regular libc memcpy/memmove calls.
 func (b *builder) createMemoryCopyImpl() {
 	b.createFunctionStart(true)
-	fnName := "llvm." + b.fn.Name() + ".p0.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+	params := b.fn.Params[0:3]
+	b.createMemCopy(
+		b.fn.Name(),
+		b.getValue(params[0], getPos(b.fn)),
+		b.getValue(params[1], getPos(b.fn)),
+		b.getValue(params[2], getPos(b.fn)),
+	)
+	b.CreateRetVoid()
+}
+
+func (b *builder) createMemCopy(kind string, dst, src, len llvm.Value) llvm.Value {
+	fnName := "llvm." + kind + ".p0.p0.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)
 		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
 	}
-	var params []llvm.Value
-	for _, param := range b.fn.Params {
-		params = append(params, b.getValue(param, getPos(b.fn)))
-	}
-	params = append(params, llvm.ConstInt(b.ctx.Int1Type(), 0, false))
-	b.CreateCall(llvmFn.GlobalValueType(), llvmFn, params, "")
-	b.CreateRetVoid()
+	return b.CreateCall(llvmFn.GlobalValueType(), llvmFn, []llvm.Value{dst, src, len, llvm.ConstInt(b.ctx.Int1Type(), 0, false)}, "")
 }
 
 // createMemoryZeroImpl creates calls to llvm.memset.* to zero a block of
@@ -144,6 +151,20 @@ func (b *builder) createAbiEscapeImpl() {
 	b.CreateRet(result)
 }
 
+// Implement machine.keepAliveNoEscape, which makes sure the compiler keeps the
+// pointer parameter alive until this point (for GC).
+func (b *builder) createMachineKeepAliveImpl() {
+	b.createFunctionStart(true)
+	pointerValue := b.getValue(b.fn.Params[0], getPos(b.fn))
+
+	// See createKeepAliveImpl for details.
+	asmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType}, false)
+	asmFn := llvm.InlineAsm(asmType, "", "r", true, false, 0, false)
+	b.createCall(asmType, asmFn, []llvm.Value{pointerValue}, "")
+
+	b.CreateRetVoid()
+}
+
 var mathToLLVMMapping = map[string]string{
 	"math.Ceil":  "llvm.ceil.f64",
 	"math.Exp":   "llvm.exp.f64",

compiler/llvm.go

@@ -1,10 +1,10 @@
 package compiler
 
 import (
+	"encoding/binary"
 	"fmt"
 	"go/token"
 	"go/types"
-	"math/big"
 	"strings"
 
 	"github.com/tinygo-org/tinygo/compileopts"
@@ -231,6 +231,12 @@ func (c *compilerContext) makeGlobalArray(buf []byte, name string, elementType l
 //
 // For details on what's in this value, see src/runtime/gc_precise.go.
 func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Value {
+	if !typeHasPointers(t) {
+		// There are no pointers in this type, so we can simplify the layout.
+		layout := (uint64(1) << 1) | 1
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
+	}
+
 	// Use the element type for arrays. This works even for nested arrays.
 	for {
 		kind := t.TypeKind()
@@ -248,54 +254,29 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		break
 	}
 
-	// Do a few checks to see whether we need to generate any object layout
-	// information at all.
+	// Create the pointer bitmap.
 	objectSizeBytes := c.targetData.TypeAllocSize(t)
+	pointerAlignment := uint64(c.targetData.PrefTypeAlignment(c.dataPtrType))
+	bitmapLen := objectSizeBytes / pointerAlignment
+	bitmapBytes := (bitmapLen + 7) / 8
+	bitmap := make([]byte, bitmapBytes, max(bitmapBytes, 8))
+	c.buildPointerBitmap(bitmap, pointerAlignment, pos, t, 0)
+
+	// Try to encode the layout inline.
 	pointerSize := c.targetData.TypeAllocSize(c.dataPtrType)
-	pointerAlignment := c.targetData.PrefTypeAlignment(c.dataPtrType)
-	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)
-	}
-	bitmap := c.getPointerBitmap(t, pos)
-	if bitmap.BitLen() == 0 {
-		// There are no pointers in this type, so we can simplify the layout.
-		// TODO: this can be done in many other cases, e.g. when allocating an
-		// array (like [4][]byte, which repeats a slice 4 times).
-		layout := (uint64(1) << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
-	}
-	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)
-	}
-	objectSizeWords := objectSizeBytes / uint64(pointerAlignment)
-
 	pointerBits := pointerSize * 8
-	var sizeFieldBits uint64
-	switch pointerBits {
-	case 16:
-		sizeFieldBits = 4
-	case 32:
-		sizeFieldBits = 5
-	case 64:
-		sizeFieldBits = 6
-	default:
-		panic("unknown pointer size")
-	}
-	layoutFieldBits := pointerBits - 1 - sizeFieldBits
+	if bitmapLen < pointerBits {
+		rawMask := binary.LittleEndian.Uint64(bitmap[0:8])
+		layout := rawMask*pointerBits + bitmapLen
+		layout <<= 1
+		layout |= 1
 
-	// Try to emit the value as an inline integer. This is possible in most
-	// cases.
-	if objectSizeWords < layoutFieldBits {
-		// If it can be stored directly in the pointer value, do so.
-		// The runtime knows that if the least significant bit of the pointer is
-		// set, the pointer contains the value itself.
-		layout := bitmap.Uint64()<<(sizeFieldBits+1) | (objectSizeWords << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
+		// Check if the layout fits.
+		layout &= 1<<pointerBits - 1
+		if (layout>>1)/pointerBits == rawMask {
+			// No set bits were shifted off.
+			return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
+		}
 	}
 
 	// Unfortunately, the object layout is too big to fit in a pointer-sized
@@ -303,25 +284,24 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 
 	// Try first whether the global already exists. All objects with a
 	// particular name have the same type, so this is possible.
-	globalName := "runtime/gc.layout:" + fmt.Sprintf("%d-%0*x", objectSizeWords, (objectSizeWords+15)/16, bitmap)
+	globalName := "runtime/gc.layout:" + fmt.Sprintf("%d-%0*x", bitmapLen, (bitmapLen+15)/16, bitmap)
 	global := c.mod.NamedGlobal(globalName)
 	if !global.IsNil() {
 		return global
 	}
 
 	// Create the global initializer.
-	bitmapBytes := make([]byte, int(objectSizeWords+7)/8)
-	bitmap.FillBytes(bitmapBytes)
-	reverseBytes(bitmapBytes) // big-endian to little-endian
-	var bitmapByteValues []llvm.Value
-	for _, b := range bitmapBytes {
-		bitmapByteValues = append(bitmapByteValues, llvm.ConstInt(c.ctx.Int8Type(), uint64(b), false))
+	bitmapByteValues := make([]llvm.Value, bitmapBytes)
+	i8 := c.ctx.Int8Type()
+	for i, b := range bitmap {
+		bitmapByteValues[i] = llvm.ConstInt(i8, uint64(b), false)
 	}
 	initializer := c.ctx.ConstStruct([]llvm.Value{
-		llvm.ConstInt(c.uintptrType, objectSizeWords, false),
-		llvm.ConstArray(c.ctx.Int8Type(), bitmapByteValues),
+		llvm.ConstInt(c.uintptrType, bitmapLen, false),
+		llvm.ConstArray(i8, bitmapByteValues),
 	}, false)
 
+	// Create the actual global.
 	global = llvm.AddGlobal(c.mod, initializer.Type(), globalName)
 	global.SetInitializer(initializer)
 	global.SetUnnamedAddr(true)
@@ -329,6 +309,7 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 	global.SetLinkage(llvm.LinkOnceODRLinkage)
 	if c.targetData.PrefTypeAlignment(c.uintptrType) < 2 {
 		// AVR doesn't have alignment by default.
+		// The lowest bit must be unset to distinguish this from an inline layout.
 		global.SetAlignment(2)
 	}
 	if c.Debug && pos != token.NoPos {
@@ -360,52 +341,71 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 	return global
 }
 
-// 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)
-	switch typ.TypeKind() {
+// buildPointerBitmap scans the given LLVM type for pointers and sets bits in a
+// bitmap at the word offset that contains a pointer. This scan is recursive.
+func (c *compilerContext) buildPointerBitmap(
+	dst []byte,
+	ptrAlign uint64,
+	pos token.Pos,
+	t llvm.Type,
+	offset uint64,
+) {
+	switch t.TypeKind() {
 	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
-		return big.NewInt(0)
+		// These types do not contain pointers.
+
 	case llvm.PointerTypeKind:
-		return big.NewInt(1)
+		// Set the corresponding position in the bitmap.
+		dst[offset/8] |= 1 << (offset % 8)
+
 	case llvm.StructTypeKind:
-		ptrs := big.NewInt(0)
-		for i, subtyp := range typ.StructElementTypes() {
-			subptrs := c.getPointerBitmap(subtyp, pos)
-			if subptrs.BitLen() == 0 {
+		// Recurse over struct elements.
+		for i, et := range t.StructElementTypes() {
+			eo := c.targetData.ElementOffset(t, i)
+			if eo%uint64(ptrAlign) != 0 {
+				if typeHasPointers(et) {
+					// This error will let the compilation fail, but by continuing
+					// the error can still easily be shown.
+					c.addError(pos, "internal error: allocated struct contains unaligned pointer")
+				}
 				continue
 			}
-			offset := c.targetData.ElementOffset(typ, i)
-			if offset%uint64(alignment) != 0 {
-				// This error will let the compilation fail, but by continuing
-				// the error can still easily be shown.
-				c.addError(pos, "internal error: allocated struct contains unaligned pointer")
-				continue
-			}
-			subptrs.Lsh(subptrs, uint(offset)/uint(alignment))
-			ptrs.Or(ptrs, subptrs)
+			c.buildPointerBitmap(
+				dst,
+				ptrAlign,
+				pos,
+				et,
+				offset+(eo/ptrAlign),
+			)
 		}
-		return ptrs
+
 	case llvm.ArrayTypeKind:
-		subtyp := typ.ElementType()
-		subptrs := c.getPointerBitmap(subtyp, pos)
-		ptrs := big.NewInt(0)
-		if subptrs.BitLen() == 0 {
-			return ptrs
+		// Recurse over array elements.
+		len := t.ArrayLength()
+		if len <= 0 {
+			return
 		}
-		elementSize := c.targetData.TypeAllocSize(subtyp)
-		if elementSize%uint64(alignment) != 0 {
-			// This error will let the compilation fail (but continues so that
-			// other errors can be shown).
-			c.addError(pos, "internal error: allocated array contains unaligned pointer")
-			return ptrs
+		et := t.ElementType()
+		elementSize := c.targetData.TypeAllocSize(et)
+		if elementSize%ptrAlign != 0 {
+			if typeHasPointers(et) {
+				// This error will let the compilation fail (but continues so that
+				// other errors can be shown).
+				c.addError(pos, "internal error: allocated array contains unaligned pointer")
+			}
+			return
 		}
-		for i := 0; i < typ.ArrayLength(); i++ {
-			ptrs.Lsh(ptrs, uint(elementSize)/uint(alignment))
-			ptrs.Or(ptrs, subptrs)
+		elementSize /= ptrAlign
+		for i := 0; i < len; i++ {
+			c.buildPointerBitmap(
+				dst,
+				ptrAlign,
+				pos,
+				et,
+				offset+uint64(i)*elementSize,
+			)
 		}
-		return ptrs
+
 	default:
 		// Should not happen.
 		panic("unknown LLVM type")

compiler/map.go

@@ -250,6 +250,9 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 func hashmapIsBinaryKey(keyType types.Type) bool {
 	switch keyType := keyType.Underlying().(type) {
 	case *types.Basic:
+		// TODO: unsafe.Pointer is also a binary key, but to support that we
+		// need to fix an issue with interp first (see
+		// https://github.com/tinygo-org/tinygo/pull/4898).
 		return keyType.Info()&(types.IsBoolean|types.IsInteger) != 0
 	case *types.Pointer:
 		return true

compiler/symbol.go

@@ -142,6 +142,11 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 			nocapture := c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0)
 			llvmFn.AddAttributeAtIndex(i+1, nocapture)
 		}
+		if paramInfo.flags&paramIsReadonly != 0 && paramInfo.llvmType.TypeKind() == llvm.PointerTypeKind {
+			// Readonly pointer parameters (like strings) benefit from being marked as readonly.
+			readonly := c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0)
+			llvmFn.AddAttributeAtIndex(i+1, readonly)
+		}
 	}
 
 	// Set a number of function or parameter attributes, depending on the
@@ -154,6 +159,8 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("noreturn"), 0))
 	case "internal/abi.NoEscape":
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+	case "machine.keepAliveNoEscape", "machine.unsafeNoEscape":
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 	case "runtime.alloc":
 		// Tell the optimizer that runtime.alloc is an allocator, meaning that it
 		// returns values that are never null and never alias to an existing value.
@@ -177,12 +184,6 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		// be modified.
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
-	case "runtime.sliceCopy":
-		// Copying a slice won't capture any of the parameters.
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("writeonly"), 0))
-		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
-		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
-		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 	case "runtime.stringFromBytes":
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))

compiler/syscall.go

@@ -74,6 +74,14 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		return b.CreateCall(fnType, target, args, ""), nil
 
 	case b.GOARCH == "arm" && b.GOOS == "linux":
+		if arch := b.archFamily(); arch != "arm" {
+			// Some targets pretend to be linux/arm for compatibility but aren't
+			// actually such a system. Make sure we emit an error instead of
+			// creating inline assembly that will fail to compile.
+			// See: https://github.com/tinygo-org/tinygo/issues/4959
+			return llvm.Value{}, b.makeError(call.Pos(), "system calls are not supported: target emulates a linux/arm system on "+arch)
+		}
+
 		// Implement the EABI system call convention for Linux.
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}

compiler/testdata/defer-cortex-m-qemu.ll

@@ -270,6 +270,241 @@ entry:
   ret void
 }
 
+; Function Attrs: nounwind
+define hidden void @main.deferInfiniteLoop(ptr %context) unnamed_addr #1 {
+entry:
+  %deferPtr = alloca ptr, align 4
+  store ptr null, ptr %deferPtr, align 4
+  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %0 = call ptr @llvm.stacksave.p0()
+  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %defer.next = load ptr, ptr %deferPtr, align 4
+  %defer.alloc.call = call dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #4
+  store i32 0, ptr %defer.alloc.call, align 4
+  %defer.alloc.call.repack1 = getelementptr inbounds nuw i8, ptr %defer.alloc.call, i32 4
+  store ptr %defer.next, ptr %defer.alloc.call.repack1, align 4
+  %defer.alloc.call.repack3 = getelementptr inbounds nuw i8, ptr %defer.alloc.call, i32 8
+  store i32 8, ptr %defer.alloc.call.repack3, align 4
+  store ptr %defer.alloc.call, ptr %deferPtr, align 4
+  br label %for.body
+
+recover:                                          ; preds = %rundefers.end
+  ret void
+
+lpad:                                             ; No predecessors!
+  br label %rundefers.loophead
+
+rundefers.loophead:                               ; preds = %rundefers.callback0, %lpad
+  br i1 poison, label %rundefers.end, label %rundefers.loop
+
+rundefers.loop:                                   ; preds = %rundefers.loophead
+  switch i32 poison, label %rundefers.default [
+    i32 0, label %rundefers.callback0
+  ]
+
+rundefers.callback0:                              ; preds = %rundefers.loop
+  br label %rundefers.loophead
+
+rundefers.default:                                ; preds = %rundefers.loop
+  unreachable
+
+rundefers.end:                                    ; preds = %rundefers.loophead
+  br label %recover
+}
+
+; Function Attrs: nounwind
+define hidden void @main.deferLoop(ptr %context) unnamed_addr #1 {
+entry:
+  %deferPtr = alloca ptr, align 4
+  store ptr null, ptr %deferPtr, align 4
+  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %0 = call ptr @llvm.stacksave.p0()
+  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
+  br label %for.loop
+
+for.loop:                                         ; preds = %for.body, %entry
+  %1 = phi i32 [ 0, %entry ], [ %3, %for.body ]
+  %2 = icmp slt i32 %1, 10
+  br i1 %2, label %for.body, label %for.done
+
+for.body:                                         ; preds = %for.loop
+  %defer.next = load ptr, ptr %deferPtr, align 4
+  %defer.alloc.call = call dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #4
+  store i32 0, ptr %defer.alloc.call, align 4
+  %defer.alloc.call.repack13 = getelementptr inbounds nuw i8, ptr %defer.alloc.call, i32 4
+  store ptr %defer.next, ptr %defer.alloc.call.repack13, align 4
+  %defer.alloc.call.repack15 = getelementptr inbounds nuw i8, ptr %defer.alloc.call, i32 8
+  store i32 %1, ptr %defer.alloc.call.repack15, align 4
+  store ptr %defer.alloc.call, ptr %deferPtr, align 4
+  %3 = add i32 %1, 1
+  br label %for.loop
+
+for.done:                                         ; preds = %for.loop
+  br label %rundefers.block
+
+rundefers.after:                                  ; preds = %rundefers.end
+  call void @runtime.destroyDeferFrame(ptr nonnull %deferframe.buf, ptr undef) #4
+  ret void
+
+rundefers.block:                                  ; preds = %for.done
+  br label %rundefers.loophead
+
+rundefers.loophead:                               ; preds = %rundefers.callback0, %rundefers.block
+  %4 = load ptr, ptr %deferPtr, align 4
+  %stackIsNil = icmp eq ptr %4, null
+  br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
+
+rundefers.loop:                                   ; preds = %rundefers.loophead
+  %stack.next.gep = getelementptr inbounds nuw i8, ptr %4, i32 4
+  %stack.next = load ptr, ptr %stack.next.gep, align 4
+  store ptr %stack.next, ptr %deferPtr, align 4
+  %callback = load i32, ptr %4, align 4
+  switch i32 %callback, label %rundefers.default [
+    i32 0, label %rundefers.callback0
+  ]
+
+rundefers.callback0:                              ; preds = %rundefers.loop
+  %gep = getelementptr inbounds nuw i8, ptr %4, i32 8
+  %param = load i32, ptr %gep, align 4
+  call void @runtime.printlock(ptr undef) #4
+  call void @runtime.printint32(i32 %param, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
+  br label %rundefers.loophead
+
+rundefers.default:                                ; preds = %rundefers.loop
+  unreachable
+
+rundefers.end:                                    ; preds = %rundefers.loophead
+  br label %rundefers.after
+
+recover:                                          ; preds = %rundefers.end1
+  ret void
+
+lpad:                                             ; No predecessors!
+  br label %rundefers.loophead4
+
+rundefers.loophead4:                              ; preds = %rundefers.callback010, %lpad
+  br i1 poison, label %rundefers.end1, label %rundefers.loop3
+
+rundefers.loop3:                                  ; preds = %rundefers.loophead4
+  switch i32 poison, label %rundefers.default2 [
+    i32 0, label %rundefers.callback010
+  ]
+
+rundefers.callback010:                            ; preds = %rundefers.loop3
+  br label %rundefers.loophead4
+
+rundefers.default2:                               ; preds = %rundefers.loop3
+  unreachable
+
+rundefers.end1:                                   ; preds = %rundefers.loophead4
+  br label %recover
+}
+
+; Function Attrs: nounwind
+define hidden void @main.deferBetweenLoops(ptr %context) unnamed_addr #1 {
+entry:
+  %defer.alloca = alloca { i32, ptr, i32 }, align 4
+  %deferPtr = alloca ptr, align 4
+  store ptr null, ptr %deferPtr, align 4
+  %deferframe.buf = alloca %runtime.deferFrame, align 4
+  %0 = call ptr @llvm.stacksave.p0()
+  call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
+  br label %for.loop
+
+for.loop:                                         ; preds = %for.body, %entry
+  %1 = phi i32 [ 0, %entry ], [ %3, %for.body ]
+  %2 = icmp slt i32 %1, 10
+  br i1 %2, label %for.body, label %for.done
+
+for.body:                                         ; preds = %for.loop
+  %3 = add i32 %1, 1
+  br label %for.loop
+
+for.done:                                         ; preds = %for.loop
+  %defer.next = load ptr, ptr %deferPtr, align 4
+  store i32 0, ptr %defer.alloca, align 4
+  %defer.alloca.repack16 = getelementptr inbounds nuw i8, ptr %defer.alloca, i32 4
+  store ptr %defer.next, ptr %defer.alloca.repack16, align 4
+  %defer.alloca.repack18 = getelementptr inbounds nuw i8, ptr %defer.alloca, i32 8
+  store i32 1, ptr %defer.alloca.repack18, align 4
+  store ptr %defer.alloca, ptr %deferPtr, align 4
+  br label %for.loop1
+
+for.loop1:                                        ; preds = %for.body2, %for.done
+  %4 = phi i32 [ 0, %for.done ], [ %6, %for.body2 ]
+  %5 = icmp slt i32 %4, 10
+  br i1 %5, label %for.body2, label %for.done3
+
+for.body2:                                        ; preds = %for.loop1
+  %6 = add i32 %4, 1
+  br label %for.loop1
+
+for.done3:                                        ; preds = %for.loop1
+  br label %rundefers.block
+
+rundefers.after:                                  ; preds = %rundefers.end
+  call void @runtime.destroyDeferFrame(ptr nonnull %deferframe.buf, ptr undef) #4
+  ret void
+
+rundefers.block:                                  ; preds = %for.done3
+  br label %rundefers.loophead
+
+rundefers.loophead:                               ; preds = %rundefers.callback0, %rundefers.block
+  %7 = load ptr, ptr %deferPtr, align 4
+  %stackIsNil = icmp eq ptr %7, null
+  br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
+
+rundefers.loop:                                   ; preds = %rundefers.loophead
+  %stack.next.gep = getelementptr inbounds nuw i8, ptr %7, i32 4
+  %stack.next = load ptr, ptr %stack.next.gep, align 4
+  store ptr %stack.next, ptr %deferPtr, align 4
+  %callback = load i32, ptr %7, align 4
+  switch i32 %callback, label %rundefers.default [
+    i32 0, label %rundefers.callback0
+  ]
+
+rundefers.callback0:                              ; preds = %rundefers.loop
+  %gep = getelementptr inbounds nuw i8, ptr %7, i32 8
+  %param = load i32, ptr %gep, align 4
+  call void @runtime.printlock(ptr undef) #4
+  call void @runtime.printint32(i32 %param, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
+  br label %rundefers.loophead
+
+rundefers.default:                                ; preds = %rundefers.loop
+  unreachable
+
+rundefers.end:                                    ; preds = %rundefers.loophead
+  br label %rundefers.after
+
+recover:                                          ; preds = %rundefers.end4
+  ret void
+
+lpad:                                             ; No predecessors!
+  br label %rundefers.loophead7
+
+rundefers.loophead7:                              ; preds = %rundefers.callback013, %lpad
+  br i1 poison, label %rundefers.end4, label %rundefers.loop6
+
+rundefers.loop6:                                  ; preds = %rundefers.loophead7
+  switch i32 poison, label %rundefers.default5 [
+    i32 0, label %rundefers.callback013
+  ]
+
+rundefers.callback013:                            ; preds = %rundefers.loop6
+  br label %rundefers.loophead7
+
+rundefers.default5:                               ; preds = %rundefers.loop6
+  unreachable
+
+rundefers.end4:                                   ; preds = %rundefers.loophead7
+  br label %recover
+}
+
 attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
 attributes #1 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
 attributes #2 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }

compiler/testdata/defer.go

@@ -18,3 +18,23 @@ func deferMultiple() {
 	}()
 	external()
 }
+
+func deferInfiniteLoop() {
+	for {
+		defer print(8)
+	}
+}
+
+func deferLoop() {
+	for i := 0; i < 10; i++ {
+		defer print(i)
+	}
+}
+
+func deferBetweenLoops() {
+	for i := 0; i < 10; i++ {
+	}
+	defer print(1)
+	for i := 0; i < 10; i++ {
+	}
+}

compiler/testdata/gc.go

@@ -24,6 +24,10 @@ var (
 		x *byte
 		y [61]uintptr
 	}
+	struct5 *struct {
+		x *byte
+		y [30]uintptr
+	}
 
 	slice1 []byte
 	slice2 []*int
@@ -58,6 +62,10 @@ func newStruct() {
 		x *byte
 		y [61]uintptr
 	})
+	struct5 = new(struct {
+		x *byte
+		y [30]uintptr
+	})
 }
 
 func newFuncValue() *func() {

compiler/testdata/gc.ll

@@ -16,11 +16,12 @@ target triple = "wasm32-unknown-wasi"
 @main.struct2 = hidden global ptr null, align 4
 @main.struct3 = hidden global ptr null, align 4
 @main.struct4 = hidden global ptr null, align 4
+@main.struct5 = hidden global ptr null, align 4
 @main.slice1 = hidden global { ptr, i32, i32 } zeroinitializer, align 4
 @main.slice2 = hidden global { ptr, i32, i32 } zeroinitializer, align 4
 @main.slice3 = hidden global { ptr, i32, i32 } zeroinitializer, align 4
-@"runtime/gc.layout:62-2000000000000001" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c"\01\00\00\00\00\00\00 " }
-@"runtime/gc.layout:62-0001" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c"\01\00\00\00\00\00\00\00" }
+@"runtime/gc.layout:62-0100000000000020" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c"\01\00\00\00\00\00\00 " }
+@"runtime/gc.layout:62-0100000000000000" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c"\01\00\00\00\00\00\00\00" }
 @"reflect/types.type:basic:complex128" = linkonce_odr constant { i8, ptr } { i8 80, ptr @"reflect/types.type:pointer:basic:complex128" }, align 4
 @"reflect/types.type:pointer:basic:complex128" = linkonce_odr constant { i8, i16, ptr } { i8 -43, i16 0, ptr @"reflect/types.type:basic:complex128" }, align 4
 
@@ -80,12 +81,15 @@ entry:
   %new1 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new1, ptr @main.struct2, align 4
-  %new2 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-2000000000000001", ptr undef) #3
+  %new2 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0100000000000020", ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new2, ptr @main.struct3, align 4
-  %new3 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0001", ptr undef) #3
+  %new3 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0100000000000000", ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new3, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new3, ptr @main.struct4, align 4
+  %new4 = call align 4 dereferenceable(124) ptr @runtime.alloc(i32 124, ptr nonnull inttoptr (i32 127 to ptr), ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %new4, ptr nonnull %stackalloc, ptr undef) #3
+  store ptr %new4, ptr @main.struct5, align 4
   ret void
 }
 

compiler/testdata/go1.20.ll

@@ -50,7 +50,7 @@ unsafe.String.throw:                              ; preds = %entry
 declare void @runtime.unsafeSlicePanic(ptr) #1
 
 ; Function Attrs: nounwind
-define hidden ptr @main.unsafeStringData(ptr %s.data, i32 %s.len, ptr %context) unnamed_addr #2 {
+define hidden ptr @main.unsafeStringData(ptr readonly %s.data, i32 %s.len, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
   call void @runtime.trackPointer(ptr %s.data, ptr nonnull %stackalloc, ptr undef) #3

compiler/testdata/go1.21.ll

@@ -22,6 +22,9 @@ entry:
   ret i32 %a
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.smin.i32(i32, i32) #3
+
 ; Function Attrs: nounwind
 define hidden i32 @main.min2(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
 entry:
@@ -53,6 +56,9 @@ entry:
   ret i8 %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i8 @llvm.umin.i8(i8, i8) #3
+
 ; Function Attrs: nounwind
 define hidden i32 @main.minUnsigned(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
 entry:
@@ -60,24 +66,31 @@ entry:
   ret i32 %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #3
+
 ; Function Attrs: nounwind
 define hidden float @main.minFloat32(float %a, float %b, ptr %context) unnamed_addr #2 {
 entry:
-  %0 = fcmp olt float %a, %b
-  %1 = select i1 %0, float %a, float %b
-  ret float %1
+  %0 = call float @llvm.minimum.f32(float %a, float %b)
+  ret float %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.minimum.f32(float, float) #3
+
 ; Function Attrs: nounwind
 define hidden double @main.minFloat64(double %a, double %b, ptr %context) unnamed_addr #2 {
 entry:
-  %0 = fcmp olt double %a, %b
-  %1 = select i1 %0, double %a, double %b
-  ret double %1
+  %0 = call double @llvm.minimum.f64(double %a, double %b)
+  ret double %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare double @llvm.minimum.f64(double, double) #3
+
 ; Function Attrs: nounwind
-define hidden %runtime._string @main.minString(ptr %a.data, i32 %a.len, ptr %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.minString(ptr readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
 entry:
   %0 = insertvalue %runtime._string zeroinitializer, ptr %a.data, 0
   %1 = insertvalue %runtime._string %0, i32 %a.len, 1
@@ -91,7 +104,7 @@ entry:
   ret %runtime._string %5
 }
 
-declare i1 @runtime.stringLess(ptr, i32, ptr, i32, ptr) #1
+declare i1 @runtime.stringLess(ptr readonly, i32, ptr readonly, i32, ptr) #1
 
 ; Function Attrs: nounwind
 define hidden i32 @main.maxInt(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
@@ -100,6 +113,9 @@ entry:
   ret i32 %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.smax.i32(i32, i32) #3
+
 ; Function Attrs: nounwind
 define hidden i32 @main.maxUint(i32 %a, i32 %b, ptr %context) unnamed_addr #2 {
 entry:
@@ -107,16 +123,21 @@ entry:
   ret i32 %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umax.i32(i32, i32) #3
+
 ; Function Attrs: nounwind
 define hidden float @main.maxFloat32(float %a, float %b, ptr %context) unnamed_addr #2 {
 entry:
-  %0 = fcmp ogt float %a, %b
-  %1 = select i1 %0, float %a, float %b
-  ret float %1
+  %0 = call float @llvm.maximum.f32(float %a, float %b)
+  ret float %0
 }
 
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare float @llvm.maximum.f32(float, float) #3
+
 ; Function Attrs: nounwind
-define hidden %runtime._string @main.maxString(ptr %a.data, i32 %a.len, ptr %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
+define hidden %runtime._string @main.maxString(ptr readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
 entry:
   %0 = insertvalue %runtime._string zeroinitializer, ptr %a.data, 0
   %1 = insertvalue %runtime._string %0, i32 %a.len, 1
@@ -139,7 +160,7 @@ entry:
 }
 
 ; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
-declare void @llvm.memset.p0.i32(ptr nocapture writeonly, i8, i32, i1 immarg) #3
+declare void @llvm.memset.p0.i32(ptr nocapture writeonly, i8, i32, i1 immarg) #4
 
 ; Function Attrs: nounwind
 define hidden void @main.clearZeroSizedSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
@@ -156,24 +177,9 @@ entry:
 
 declare void @runtime.hashmapClear(ptr dereferenceable_or_null(40), ptr) #1
 
-; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.smin.i32(i32, i32) #4
-
-; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i8 @llvm.umin.i8(i8, i8) #4
-
-; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.umin.i32(i32, i32) #4
-
-; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.smax.i32(i32, i32) #4
-
-; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
-declare i32 @llvm.umax.i32(i32, i32) #4
-
 attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nocallback nofree nounwind willreturn memory(argmem: write) }
-attributes #4 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #3 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: write) }
 attributes #5 = { nounwind }

compiler/testdata/goroutine-cortex-m-qemu-tasks.ll

@@ -17,8 +17,8 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr undef) #9
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr undef) #11
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #11
   ret void
 }
 
@@ -28,7 +28,7 @@ declare void @main.regularFunction(i32, ptr) #2
 define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #3 {
 entry:
   %unpack.int = ptrtoint ptr %0 to i32
-  call void @main.regularFunction(i32 %unpack.int, ptr undef) #9
+  call void @main.regularFunction(i32 %unpack.int, ptr undef) #11
   ret void
 }
 
@@ -39,8 +39,8 @@ declare void @"internal/task.start"(i32, ptr, i32, ptr) #2
 ; Function Attrs: nounwind
 define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr undef) #9
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr undef) #11
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 %stacksize, ptr undef) #11
   ret void
 }
 
@@ -61,18 +61,18 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #1 {
 entry:
-  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
+  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #11
   store i32 3, ptr %n, align 4
-  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #11
   store i32 5, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr %n, ptr %1, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr undef) #9
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr undef) #11
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #11
   %2 = load i32, ptr %n, align 4
-  call void @runtime.printlock(ptr undef) #9
-  call void @runtime.printint32(i32 %2, ptr undef) #9
-  call void @runtime.printunlock(ptr undef) #9
+  call void @runtime.printlock(ptr undef) #11
+  call void @runtime.printint32(i32 %2, ptr undef) #11
+  call void @runtime.printunlock(ptr undef) #11
   ret void
 }
 
@@ -102,14 +102,14 @@ declare void @runtime.printunlock(ptr) #2
 ; Function Attrs: nounwind
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #11
   store i32 5, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr %fn.context, ptr %1, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 8
   store ptr %fn.funcptr, ptr %2, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr undef) #9
-  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr undef) #11
+  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #11
   ret void
 }
 
@@ -121,7 +121,7 @@ entry:
   %3 = load ptr, ptr %2, align 4
   %4 = getelementptr inbounds nuw i8, ptr %0, i32 8
   %5 = load ptr, ptr %4, align 4
-  call void %5(i32 %1, ptr %3) #9
+  call void %5(i32 %1, ptr %3) #11
   ret void
 }
 
@@ -134,16 +134,21 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.copyBuiltinGoroutine(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #1 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(ptr %dst.data, ptr %src.data, i32 %dst.len, i32 %src.len, i32 1, ptr undef) #9
+  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
+  call void @llvm.memmove.p0.p0.i32(ptr align 1 %dst.data, ptr align 1 %src.data, i32 %copy.n, i1 false)
   ret void
 }
 
-declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #2
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #7
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.memmove.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #8
 
 ; Function Attrs: nounwind
 define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
-  call void @runtime.chanClose(ptr %ch, ptr undef) #9
+  call void @runtime.chanClose(ptr %ch, ptr undef) #11
   ret void
 }
 
@@ -152,7 +157,7 @@ declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #2
 ; Function Attrs: nounwind
 define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
 entry:
-  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #9
+  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #11
   store ptr %itf.value, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr @"main$string", ptr %1, align 4
@@ -160,15 +165,15 @@ entry:
   store i32 4, ptr %2, align 4
   %3 = getelementptr inbounds nuw i8, ptr %0, i32 12
   store ptr %itf.typecode, ptr %3, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr undef) #9
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr undef) #11
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #11
   ret void
 }
 
-declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #7
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #9
 
 ; Function Attrs: nounwind
-define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #8 {
+define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #10 {
 entry:
   %1 = load ptr, ptr %0, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -177,7 +182,7 @@ entry:
   %5 = load i32, ptr %4, align 4
   %6 = getelementptr inbounds nuw i8, ptr %0, i32 12
   %7 = load ptr, ptr %6, align 4
-  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #9
+  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #11
   ret void
 }
 
@@ -188,6 +193,8 @@ attributes #3 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb
 attributes #4 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
 attributes #5 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
 attributes #6 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper" }
-attributes #7 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
-attributes #8 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
-attributes #9 = { nounwind }
+attributes #7 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #8 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+attributes #9 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #10 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
+attributes #11 = { nounwind }

compiler/testdata/goroutine-wasm-asyncify.ll

@@ -19,7 +19,7 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.regularFunctionGoroutine(ptr %context) unnamed_addr #2 {
 entry:
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.regularFunction$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #11
   ret void
 }
 
@@ -31,8 +31,8 @@ declare void @runtime.deadlock(ptr) #1
 define linkonce_odr void @"main.regularFunction$gowrapper"(ptr %0) unnamed_addr #3 {
 entry:
   %unpack.int = ptrtoint ptr %0 to i32
-  call void @main.regularFunction(i32 %unpack.int, ptr undef) #9
-  call void @runtime.deadlock(ptr undef) #9
+  call void @main.regularFunction(i32 %unpack.int, ptr undef) #11
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -41,7 +41,7 @@ declare void @"internal/task.start"(i32, ptr, i32, ptr) #1
 ; Function Attrs: nounwind
 define hidden void @main.inlineFunctionGoroutine(ptr %context) unnamed_addr #2 {
 entry:
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.inlineFunctionGoroutine$1$gowrapper" to i32), ptr nonnull inttoptr (i32 5 to ptr), i32 65536, ptr undef) #11
   ret void
 }
 
@@ -56,7 +56,7 @@ define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(ptr %0) unn
 entry:
   %unpack.int = ptrtoint ptr %0 to i32
   call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, ptr undef)
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -64,21 +64,21 @@ entry:
 define hidden void @main.closureFunctionGoroutine(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #9
-  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #9
+  %n = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #11
+  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #11
   store i32 3, ptr %n, align 4
-  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #9
-  call void @runtime.trackPointer(ptr nonnull @"main.closureFunctionGoroutine$1", ptr nonnull %stackalloc, ptr undef) #9
-  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #9
-  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #9
+  call void @runtime.trackPointer(ptr nonnull %n, ptr nonnull %stackalloc, ptr undef) #11
+  call void @runtime.trackPointer(ptr nonnull @"main.closureFunctionGoroutine$1", ptr nonnull %stackalloc, ptr undef) #11
+  %0 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr null, ptr undef) #11
+  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #11
   store i32 5, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr %n, ptr %1, align 4
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #11
   %2 = load i32, ptr %n, align 4
-  call void @runtime.printlock(ptr undef) #9
-  call void @runtime.printint32(i32 %2, ptr undef) #9
-  call void @runtime.printunlock(ptr undef) #9
+  call void @runtime.printlock(ptr undef) #11
+  call void @runtime.printint32(i32 %2, ptr undef) #11
+  call void @runtime.printunlock(ptr undef) #11
   ret void
 }
 
@@ -96,7 +96,7 @@ entry:
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 4
   %3 = load ptr, ptr %2, align 4
   call void @"main.closureFunctionGoroutine$1"(i32 %1, ptr %3)
-  call void @runtime.deadlock(ptr undef) #9
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -110,14 +110,14 @@ declare void @runtime.printunlock(ptr) #1
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #9
-  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #9
+  %0 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr null, ptr undef) #11
+  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #11
   store i32 5, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr %fn.context, ptr %1, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 8
   store ptr %fn.funcptr, ptr %2, align 4
-  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @main.funcGoroutine.gowrapper to i32), ptr nonnull %0, i32 65536, ptr undef) #11
   ret void
 }
 
@@ -129,8 +129,8 @@ entry:
   %3 = load ptr, ptr %2, align 4
   %4 = getelementptr inbounds nuw i8, ptr %0, i32 8
   %5 = load ptr, ptr %4, align 4
-  call void %5(i32 %1, ptr %3) #9
-  call void @runtime.deadlock(ptr undef) #9
+  call void %5(i32 %1, ptr %3) #11
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -143,16 +143,21 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.copyBuiltinGoroutine(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #2 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(ptr %dst.data, ptr %src.data, i32 %dst.len, i32 %src.len, i32 1, ptr undef) #9
+  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
+  call void @llvm.memmove.p0.p0.i32(ptr align 1 %dst.data, ptr align 1 %src.data, i32 %copy.n, i1 false)
   ret void
 }
 
-declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #1
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #7
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.memmove.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #8
 
 ; Function Attrs: nounwind
 define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  call void @runtime.chanClose(ptr %ch, ptr undef) #9
+  call void @runtime.chanClose(ptr %ch, ptr undef) #11
   ret void
 }
 
@@ -162,8 +167,8 @@ declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #1
 define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #9
-  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #9
+  %0 = call align 4 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #11
+  call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #11
   store ptr %itf.value, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %0, i32 4
   store ptr @"main$string", ptr %1, align 4
@@ -171,14 +176,14 @@ entry:
   store i32 4, ptr %2, align 4
   %3 = getelementptr inbounds nuw i8, ptr %0, i32 12
   store ptr %itf.typecode, ptr %3, align 4
-  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #9
+  call void @"internal/task.start"(i32 ptrtoint (ptr @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #11
   ret void
 }
 
-declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #7
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr, ptr, i32, ptr, ptr) #9
 
 ; Function Attrs: nounwind
-define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #8 {
+define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(ptr %0) unnamed_addr #10 {
 entry:
   %1 = load ptr, ptr %0, align 4
   %2 = getelementptr inbounds nuw i8, ptr %0, i32 4
@@ -187,8 +192,8 @@ entry:
   %5 = load i32, ptr %4, align 4
   %6 = getelementptr inbounds nuw i8, ptr %0, i32 12
   %7 = load ptr, ptr %6, align 4
-  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #9
-  call void @runtime.deadlock(ptr undef) #9
+  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(ptr %1, ptr %3, i32 %5, ptr %7, ptr undef) #11
+  call void @runtime.deadlock(ptr undef) #11
   unreachable
 }
 
@@ -199,6 +204,8 @@ attributes #3 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+cal
 attributes #4 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
 attributes #5 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
 attributes #6 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper" }
-attributes #7 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
-attributes #8 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
-attributes #9 = { nounwind }
+attributes #7 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #8 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+attributes #9 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #10 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
+attributes #11 = { nounwind }

compiler/testdata/slice.ll

@@ -38,7 +38,7 @@ lookup.next:                                      ; preds = %entry
   ret i32 %1
 
 lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(ptr undef) #3
+  call void @runtime.lookupPanic(ptr undef) #5
   unreachable
 }
 
@@ -48,49 +48,55 @@ declare void @runtime.lookupPanic(ptr) #1
 define hidden { ptr, i32, i32 } @main.sliceAppendValues(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %varargs = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
-  call void @runtime.trackPointer(ptr nonnull %varargs, ptr nonnull %stackalloc, ptr undef) #3
+  %varargs = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
+  call void @runtime.trackPointer(ptr nonnull %varargs, ptr nonnull %stackalloc, ptr undef) #5
   store i32 1, ptr %varargs, align 4
   %0 = getelementptr inbounds nuw i8, ptr %varargs, i32 4
   store i32 2, ptr %0, align 4
   %1 = getelementptr inbounds nuw i8, ptr %varargs, i32 8
   store i32 3, ptr %1, align 4
-  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr nonnull %varargs, i32 %ints.len, i32 %ints.cap, i32 3, i32 4, ptr undef) #3
+  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr nonnull %varargs, i32 %ints.len, i32 %ints.cap, i32 3, i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %append.newPtr = extractvalue { ptr, i32, i32 } %append.new, 0
   %append.newLen = extractvalue { ptr, i32, i32 } %append.new, 1
   %append.newCap = extractvalue { ptr, i32, i32 } %append.new, 2
   %2 = insertvalue { ptr, i32, i32 } undef, ptr %append.newPtr, 0
   %3 = insertvalue { ptr, i32, i32 } %2, i32 %append.newLen, 1
   %4 = insertvalue { ptr, i32, i32 } %3, i32 %append.newCap, 2
-  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %4
 }
 
-declare { ptr, i32, i32 } @runtime.sliceAppend(ptr, ptr nocapture readonly, i32, i32, i32, i32, ptr) #1
+declare { ptr, i32, i32 } @runtime.sliceAppend(ptr, ptr nocapture readonly, i32, i32, i32, i32, ptr, ptr) #1
 
 ; Function Attrs: nounwind
 define hidden { ptr, i32, i32 } @main.sliceAppendSlice(ptr %ints.data, i32 %ints.len, i32 %ints.cap, ptr %added.data, i32 %added.len, i32 %added.cap, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr %added.data, i32 %ints.len, i32 %ints.cap, i32 %added.len, i32 4, ptr undef) #3
+  %append.new = call { ptr, i32, i32 } @runtime.sliceAppend(ptr %ints.data, ptr %added.data, i32 %ints.len, i32 %ints.cap, i32 %added.len, i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %append.newPtr = extractvalue { ptr, i32, i32 } %append.new, 0
   %append.newLen = extractvalue { ptr, i32, i32 } %append.new, 1
   %append.newCap = extractvalue { ptr, i32, i32 } %append.new, 2
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %append.newPtr, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %append.newLen, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %append.newCap, 2
-  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %append.newPtr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 }
 
 ; Function Attrs: nounwind
 define hidden i32 @main.sliceCopy(ptr %dst.data, i32 %dst.len, i32 %dst.cap, ptr %src.data, i32 %src.len, i32 %src.cap, ptr %context) unnamed_addr #2 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(ptr %dst.data, ptr %src.data, i32 %dst.len, i32 %src.len, i32 4, ptr undef) #3
+  %copy.n = call i32 @llvm.umin.i32(i32 %dst.len, i32 %src.len)
+  %copy.size = shl nuw i32 %copy.n, 2
+  call void @llvm.memmove.p0.p0.i32(ptr align 4 %dst.data, ptr align 4 %src.data, i32 %copy.size, i1 false)
   ret i32 %copy.n
 }
 
-declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #1
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare i32 @llvm.umin.i32(i32, i32) #3
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.memmove.p0.p0.i32(ptr nocapture writeonly, ptr nocapture readonly, i32, i1 immarg) #4
 
 ; Function Attrs: nounwind
 define hidden { ptr, i32, i32 } @main.makeByteSlice(i32 %len, ptr %context) unnamed_addr #2 {
@@ -100,15 +106,15 @@ entry:
   br i1 %slice.maxcap, label %slice.throw, label %slice.next
 
 slice.next:                                       ; preds = %entry
-  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %len, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %len, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -123,15 +129,15 @@ entry:
 
 slice.next:                                       ; preds = %entry
   %makeslice.cap = shl nuw i32 %len, 1
-  %makeslice.buf = call align 2 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 2 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -144,15 +150,15 @@ entry:
 
 slice.next:                                       ; preds = %entry
   %makeslice.cap = mul i32 %len, 3
-  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 1 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -165,15 +171,15 @@ entry:
 
 slice.next:                                       ; preds = %entry
   %makeslice.cap = shl nuw i32 %len, 2
-  %makeslice.buf = call align 4 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice.buf = call align 4 ptr @runtime.alloc(i32 %makeslice.cap, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
   %0 = insertvalue { ptr, i32, i32 } undef, ptr %makeslice.buf, 0
   %1 = insertvalue { ptr, i32, i32 } %0, i32 %len, 1
   %2 = insertvalue { ptr, i32, i32 } %1, i32 %len, 2
-  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr nonnull %makeslice.buf, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %2
 
 slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(ptr undef) #3
+  call void @runtime.slicePanic(ptr undef) #5
   unreachable
 }
 
@@ -182,7 +188,7 @@ define hidden ptr @main.Add32(ptr %p, i32 %len, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
   %0 = getelementptr i8, ptr %p, i32 %len
-  call void @runtime.trackPointer(ptr %0, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %0, ptr nonnull %stackalloc, ptr undef) #5
   ret ptr %0
 }
 
@@ -192,7 +198,7 @@ entry:
   %stackalloc = alloca i8, align 1
   %0 = trunc i64 %len to i32
   %1 = getelementptr i8, ptr %p, i32 %0
-  call void @runtime.trackPointer(ptr %1, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %1, ptr nonnull %stackalloc, ptr undef) #5
   ret ptr %1
 }
 
@@ -206,7 +212,7 @@ slicetoarray.next:                                ; preds = %entry
   ret ptr %s.data
 
 slicetoarray.throw:                               ; preds = %entry
-  call void @runtime.sliceToArrayPointerPanic(ptr undef) #3
+  call void @runtime.sliceToArrayPointerPanic(ptr undef) #5
   unreachable
 }
 
@@ -216,8 +222,8 @@ declare void @runtime.sliceToArrayPointerPanic(ptr) #1
 define hidden ptr @main.SliceToArrayConst(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %makeslice = call align 4 dereferenceable(24) ptr @runtime.alloc(i32 24, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
-  call void @runtime.trackPointer(ptr nonnull %makeslice, ptr nonnull %stackalloc, ptr undef) #3
+  %makeslice = call align 4 dereferenceable(24) ptr @runtime.alloc(i32 24, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #5
+  call void @runtime.trackPointer(ptr nonnull %makeslice, ptr nonnull %stackalloc, ptr undef) #5
   br i1 false, label %slicetoarray.throw, label %slicetoarray.next
 
 slicetoarray.next:                                ; preds = %entry
@@ -242,11 +248,11 @@ unsafe.Slice.next:                                ; preds = %entry
   %5 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %6 = insertvalue { ptr, i32, i32 } %5, i32 %len, 1
   %7 = insertvalue { ptr, i32, i32 } %6, i32 %len, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %7
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
@@ -266,11 +272,11 @@ unsafe.Slice.next:                                ; preds = %entry
   %4 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %5 = insertvalue { ptr, i32, i32 } %4, i32 %3, 1
   %6 = insertvalue { ptr, i32, i32 } %5, i32 %3, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %6
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
@@ -290,11 +296,11 @@ unsafe.Slice.next:                                ; preds = %entry
   %6 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %7 = insertvalue { ptr, i32, i32 } %6, i32 %5, 1
   %8 = insertvalue { ptr, i32, i32 } %7, i32 %5, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %8
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
@@ -314,15 +320,17 @@ unsafe.Slice.next:                                ; preds = %entry
   %6 = insertvalue { ptr, i32, i32 } undef, ptr %ptr, 0
   %7 = insertvalue { ptr, i32, i32 } %6, i32 %5, 1
   %8 = insertvalue { ptr, i32, i32 } %7, i32 %5, 2
-  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
+  call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #5
   ret { ptr, i32, i32 } %8
 
 unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(ptr undef) #3
+  call void @runtime.unsafeSlicePanic(ptr undef) #5
   unreachable
 }
 
 attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #1 = { "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #2 = { nounwind "target-features"="+bulk-memory,+bulk-memory-opt,+call-indirect-overlong,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
-attributes #3 = { nounwind }
+attributes #3 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
+attributes #5 = { nounwind }

compiler/testdata/string.ll

@@ -31,13 +31,13 @@ entry:
 }
 
 ; Function Attrs: nounwind
-define hidden i32 @main.stringLen(ptr %s.data, i32 %s.len, ptr %context) unnamed_addr #2 {
+define hidden i32 @main.stringLen(ptr readonly %s.data, i32 %s.len, ptr %context) unnamed_addr #2 {
 entry:
   ret i32 %s.len
 }
 
 ; Function Attrs: nounwind
-define hidden i8 @main.stringIndex(ptr %s.data, i32 %s.len, i32 %index, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.stringIndex(ptr readonly %s.data, i32 %s.len, i32 %index, ptr %context) unnamed_addr #2 {
 entry:
   %.not = icmp ult i32 %index, %s.len
   br i1 %.not, label %lookup.next, label %lookup.throw
@@ -55,16 +55,16 @@ lookup.throw:                                     ; preds = %entry
 declare void @runtime.lookupPanic(ptr) #1
 
 ; Function Attrs: nounwind
-define hidden i1 @main.stringCompareEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.stringCompareEqual(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
 entry:
   %0 = call i1 @runtime.stringEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr undef) #3
   ret i1 %0
 }
 
-declare i1 @runtime.stringEqual(ptr, i32, ptr, i32, ptr) #1
+declare i1 @runtime.stringEqual(ptr readonly, i32, ptr readonly, i32, ptr) #1
 
 ; Function Attrs: nounwind
-define hidden i1 @main.stringCompareUnequal(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.stringCompareUnequal(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
 entry:
   %0 = call i1 @runtime.stringEqual(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr undef) #3
   %1 = xor i1 %0, true
@@ -72,16 +72,16 @@ entry:
 }
 
 ; Function Attrs: nounwind
-define hidden i1 @main.stringCompareLarger(ptr %s1.data, i32 %s1.len, ptr %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
+define hidden i1 @main.stringCompareLarger(ptr readonly %s1.data, i32 %s1.len, ptr readonly %s2.data, i32 %s2.len, ptr %context) unnamed_addr #2 {
 entry:
   %0 = call i1 @runtime.stringLess(ptr %s2.data, i32 %s2.len, ptr %s1.data, i32 %s1.len, ptr undef) #3
   ret i1 %0
 }
 
-declare i1 @runtime.stringLess(ptr, i32, ptr, i32, ptr) #1
+declare i1 @runtime.stringLess(ptr readonly, i32, ptr readonly, i32, ptr) #1
 
 ; Function Attrs: nounwind
-define hidden i8 @main.stringLookup(ptr %s.data, i32 %s.len, i8 %x, ptr %context) unnamed_addr #2 {
+define hidden i8 @main.stringLookup(ptr readonly %s.data, i32 %s.len, i8 %x, ptr %context) unnamed_addr #2 {
 entry:
   %0 = zext i8 %x to i32
   %.not = icmp ugt i32 %s.len, %0

flake.lock

@@ -20,16 +20,16 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1747953325,
-        "narHash": "sha256-y2ZtlIlNTuVJUZCqzZAhIw5rrKP4DOSklev6c8PyCkQ=",
+        "lastModified": 1770136044,
+        "narHash": "sha256-tlFqNG/uzz2++aAmn4v8J0vAkV3z7XngeIIB3rM3650=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "55d1f923c480dadce40f5231feb472e81b0bab48",
+        "rev": "e576e3c9cf9bad747afcddd9e34f51d18c855b4e",
         "type": "github"
       },
       "original": {
         "id": "nixpkgs",
-        "ref": "nixos-25.05",
+        "ref": "nixos-25.11",
         "type": "indirect"
       }
     },

flake.nix

@@ -34,7 +34,7 @@
   inputs = {
     # Use a recent stable release, but fix the version to make it reproducible.
     # This version should be updated from time to time.
-    nixpkgs.url = "nixpkgs/nixos-25.05";
+    nixpkgs.url = "nixpkgs/nixos-25.11";
     flake-utils.url = "github:numtide/flake-utils";
   };
   outputs = { self, nixpkgs, flake-utils }:

goenv/version.go

@@ -10,7 +10,7 @@ import (
 
 // Version of TinyGo.
 // Update this value before release of new version of software.
-const version = "0.40.0-dev"
+const version = "0.41.0-dev"
 
 // Return TinyGo version, either in the form 0.30.0 or as a development version
 // (like 0.30.0-dev-abcd012).

interp/interp_test.go

@@ -2,7 +2,6 @@ package interp
 
 import (
 	"os"
-	"strconv"
 	"strings"
 	"testing"
 	"time"
@@ -11,25 +10,17 @@ import (
 )
 
 func TestInterp(t *testing.T) {
-	llvmVersion, err := strconv.Atoi(strings.Split(llvm.Version, ".")[0])
-	if err != nil {
-		// Note: this should never happen and if it does, it will always happen
-		// for a particular build because llvm.Version is a constant.
-		panic(err)
-	}
 	for _, name := range []string{
 		"basic",
 		"phi",
-		"slice-copy",
 		"consteval",
+		"intrinsics",
+		"copy",
 		"interface",
 		"revert",
 		"alloc",
 	} {
 		name := name // make local to this closure
-		if name == "slice-copy" && llvmVersion < 14 {
-			continue
-		}
 		t.Run(name, func(t *testing.T) {
 			t.Parallel()
 			runTest(t, "testdata/"+name)

interp/interpreter.go

@@ -312,33 +312,28 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					fmt.Fprintln(os.Stderr, indent+"runtime.alloc:", size, "->", ptr)
 				}
 				locals[inst.localIndex] = ptr
-			case callFn.name == "runtime.sliceCopy":
-				// sliceCopy implements the built-in copy function for slices.
-				// It is implemented here so that it can be used even if the
-				// runtime implementation is not available. Doing it this way
-				// may also be faster.
-				// Code:
-				// func sliceCopy(dst, src unsafe.Pointer, dstLen, srcLen uintptr, elemSize uintptr) int {
-				//     n := srcLen
-				//     if n > dstLen {
-				//         n = dstLen
-				//     }
-				//     memmove(dst, src, n*elemSize)
-				//     return int(n)
-				// }
-				dstLen := operands[3].Uint(r)
-				srcLen := operands[4].Uint(r)
-				elemSize := operands[5].Uint(r)
-				n := srcLen
-				if n > dstLen {
-					n = dstLen
+			case strings.HasPrefix(callFn.name, "llvm.umin."):
+				locals[inst.localIndex] = makeLiteralInt(min(operands[1].Uint(r), operands[2].Uint(r)), inst.llvmInst.Type().IntTypeWidth())
+			case strings.HasPrefix(callFn.name, "llvm.smin."):
+				locals[inst.localIndex] = makeLiteralInt(uint64(min(operands[1].Int(r), operands[2].Int(r))), inst.llvmInst.Type().IntTypeWidth())
+			case strings.HasPrefix(callFn.name, "llvm.umax."):
+				locals[inst.localIndex] = makeLiteralInt(max(operands[1].Uint(r), operands[2].Uint(r)), inst.llvmInst.Type().IntTypeWidth())
+			case strings.HasPrefix(callFn.name, "llvm.smax."):
+				locals[inst.localIndex] = makeLiteralInt(uint64(max(operands[1].Int(r), operands[2].Int(r))), inst.llvmInst.Type().IntTypeWidth())
+			case strings.HasPrefix(callFn.name, "llvm.memcpy.p0") || strings.HasPrefix(callFn.name, "llvm.memmove.p0"):
+				// Copy a block of memory from one pointer to another.
+				if operands[4].Uint(r) != 0 {
+					// This is a volatile copy/move.
+					err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+					if err != nil {
+						return nil, mem, err
+					}
+					continue
 				}
-				if r.debug {
-					fmt.Fprintln(os.Stderr, indent+"copy:", operands[1], operands[2], n)
-				}
-				if n != 0 {
+				nBytes := operands[3].Uint(r)
+				if nBytes != 0 {
 					// Only try to copy bytes when there are any bytes to copy.
-					// This is not just an optimization. If one of the slices
+					// This is not just an optimization. If one of the pointers
 					// (or both) are nil, the asPointer method call will fail
 					// even though copying a nil slice is allowed.
 					dst, err := operands[1].asPointer(r)
@@ -363,11 +358,10 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 						}
 						continue
 					}
-					nBytes := uint32(n * elemSize)
 					srcObj := mem.get(src.index())
 					dstObj := mem.getWritable(dst.index())
 					if srcObj.buffer == nil || dstObj.buffer == nil {
-						// If the buffer is nil, it means the slice is external.
+						// If the buffer is nil, it means the memory is external.
 						// This can happen for example when copying data out of
 						// a //go:embed slice, which is not available at interp
 						// time.
@@ -380,33 +374,10 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					}
 					dstBuf := dstObj.buffer.asRawValue(r)
 					srcBuf := srcObj.buffer.asRawValue(r)
-					copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
+					copy(dstBuf.buf[dst.offset():][:nBytes], srcBuf.buf[src.offset():][:nBytes])
 					dstObj.buffer = dstBuf
 					mem.put(dst.index(), dstObj)
 				}
-				locals[inst.localIndex] = makeLiteralInt(n, inst.llvmInst.Type().IntTypeWidth())
-			case strings.HasPrefix(callFn.name, "llvm.memcpy.p0") || strings.HasPrefix(callFn.name, "llvm.memmove.p0"):
-				// Copy a block of memory from one pointer to another.
-				dst, err := operands[1].asPointer(r)
-				if err != nil {
-					return nil, mem, r.errorAt(inst, err)
-				}
-				src, err := operands[2].asPointer(r)
-				if err != nil {
-					return nil, mem, r.errorAt(inst, err)
-				}
-				nBytes := uint32(operands[3].Uint(r))
-				dstObj := mem.getWritable(dst.index())
-				dstBuf := dstObj.buffer.asRawValue(r)
-				if mem.get(src.index()).buffer == nil {
-					// Looks like the source buffer is not defined.
-					// This can happen with //extern or //go:embed.
-					return nil, mem, r.errorAt(inst, errUnsupportedRuntimeInst)
-				}
-				srcBuf := mem.get(src.index()).buffer.asRawValue(r)
-				copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
-				dstObj.buffer = dstBuf
-				mem.put(dst.index(), dstObj)
 			case callFn.name == "runtime.typeAssert":
 				// This function must be implemented manually as it is normally
 				// implemented by the interface lowering pass.
@@ -577,7 +548,7 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 						// runtime instead of at compile time. But we need to
 						// revert any changes made by the call first.
 						if r.debug {
-							fmt.Fprintln(os.Stderr, indent+"!! revert because of error:", callErr.Err)
+							fmt.Fprintln(os.Stderr, indent+"!! revert because of error:", callErr.Error())
 						}
 						callMem.revert()
 						err := r.runAtRuntime(fn, inst, locals, &mem, indent)

interp/testdata/copy.ll

@@ -0,0 +1,68 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@string = internal unnamed_addr constant [3 x i8] c"foo"
+@moveDst = global [3 x i8] zeroinitializer
+@copyDst = global [3 x i8] zeroinitializer
+
+@externalSrc = external global [2 x i8]
+@moveExternalDst = global [2 x i8] zeroinitializer
+
+@moveEscapedSrc = global [4 x i8] c"abcd"
+@moveEscapedDst = global [4 x i8] zeroinitializer
+
+@volatileSrc = global [2 x i8] c"xy"
+@volatileDst = global [2 x i8] zeroinitializer
+
+declare void @use(ptr)
+
+define void @runtime.initAll() {
+  call void @main.init()
+  ret void
+}
+
+define internal void @main.init() {
+  call void @testMove()
+  call void @testCopy()
+  call void @testMoveExternal()
+  call void @testMoveEscaped()
+  call void @testVolatileCopy()
+  ret void
+}
+
+; Test a simple memmove between globals.
+define internal void @testMove() {
+  call void @llvm.memmove.p0.p0.i64(ptr @moveDst, ptr @string, i64 3, i1 false)
+  ret void
+}
+
+; Test a simple memcpy between globals.
+define internal void @testCopy() {
+  call void @llvm.memcpy.p0.p0.i64(ptr @copyDst, ptr @string, i64 3, i1 false)
+  ret void
+}
+
+; Test a memmove from an external global.
+; This should be run at runtime.
+define internal void @testMoveExternal() {
+  call void @llvm.memmove.p0.p0.i64(ptr @moveExternalDst, ptr @externalSrc, i64 2, i1 false)
+  ret void
+}
+
+; Test a memmove from an escaped (and potentially modified) source buffer.
+define internal void @testMoveEscaped() {
+  call void @use(ptr @moveEscapedSrc)
+  call void @llvm.memmove.p0.p0.i64(ptr @moveEscapedDst, ptr @moveEscapedSrc, i64 4, i1 false)
+  ret void
+}
+
+; Test a volatile memcpy.
+; This should always be run at runtime.
+define internal void @testVolatileCopy() {
+  call void @llvm.memcpy.p0.p0.i64(ptr @volatileDst, ptr @volatileSrc, i64 2, i1 true)
+  ret void
+}
+
+declare void @llvm.memmove.p0.p0.i64(ptr, ptr, i64, i1)
+
+declare void @llvm.memcpy.p0.p0.i64(ptr, ptr, i64, i1)

interp/testdata/copy.out.ll

@@ -0,0 +1,27 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@moveDst = local_unnamed_addr global [3 x i8] c"foo"
+@copyDst = local_unnamed_addr global [3 x i8] c"foo"
+@externalSrc = external local_unnamed_addr global [2 x i8]
+@moveExternalDst = local_unnamed_addr global [2 x i8] zeroinitializer
+@moveEscapedSrc = global [4 x i8] c"abcd"
+@moveEscapedDst = local_unnamed_addr global [4 x i8] zeroinitializer
+@volatileSrc = global [2 x i8] c"xy"
+@volatileDst = global [2 x i8] zeroinitializer
+
+declare void @use(ptr) local_unnamed_addr
+
+define void @runtime.initAll() local_unnamed_addr {
+  call void @llvm.memmove.p0.p0.i64(ptr @moveExternalDst, ptr @externalSrc, i64 2, i1 false)
+  call void @use(ptr @moveEscapedSrc)
+  call void @llvm.memmove.p0.p0.i64(ptr @moveEscapedDst, ptr @moveEscapedSrc, i64 4, i1 false)
+  call void @llvm.memcpy.p0.p0.i64(ptr @volatileDst, ptr @volatileSrc, i64 2, i1 true)
+  ret void
+}
+
+declare void @llvm.memmove.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1 immarg) #0
+
+declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg) #0
+
+attributes #0 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }

interp/testdata/intrinsics.ll

@@ -0,0 +1,52 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@uminResult = global i32 0
+@sminResult = global i32 0
+@umaxResult = global i32 0
+@smaxResult = global i32 0
+
+define void @runtime.initAll() {
+  call void @main.init()
+  ret void
+}
+
+define internal void @main.init() {
+  call void @testUMin()
+  call void @testSMin()
+  call void @testUMax()
+  call void @testSMax()
+  ret void
+}
+
+define internal void @testUMin() {
+  %umin = call i32 @llvm.umin.i32(i32 12, i32 -1)
+  store i32 %umin, ptr @uminResult
+  ret void
+}
+
+declare i32 @llvm.umin.i32(i32, i32)
+
+define internal void @testSMin() {
+  %smin = call i32 @llvm.smin.i32(i32 12, i32 -1)
+  store i32 %smin, ptr @sminResult
+  ret void
+}
+
+declare i32 @llvm.smin.i32(i32, i32)
+
+define internal void @testUMax() {
+  %umax = call i32 @llvm.umax.i32(i32 12, i32 -1)
+  store i32 %umax, ptr @umaxResult
+  ret void
+}
+
+declare i32 @llvm.umax.i32(i32, i32)
+
+define internal void @testSMax() {
+  %smax = call i32 @llvm.smax.i32(i32 12, i32 -1)
+  store i32 %smax, ptr @smaxResult
+  ret void
+}
+
+declare i32 @llvm.smax.i32(i32, i32)

interp/testdata/intrinsics.out.ll

@@ -0,0 +1,11 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@uminResult = local_unnamed_addr global i32 12
+@sminResult = local_unnamed_addr global i32 -1
+@umaxResult = local_unnamed_addr global i32 -1
+@smaxResult = local_unnamed_addr global i32 12
+
+define void @runtime.initAll() local_unnamed_addr {
+  ret void
+}

interp/testdata/slice-copy.ll

@@ -1,124 +0,0 @@
-target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64--linux"
-
-@main.uint8SliceSrc.buf = internal global [2 x i8] c"\03d"
-@main.uint8SliceSrc = internal unnamed_addr global { ptr, i64, i64 } { ptr @main.uint8SliceSrc.buf, i64 2, i64 2 }
-@main.uint8SliceDst = internal unnamed_addr global { ptr, i64, i64 } zeroinitializer
-@main.int16SliceSrc.buf = internal global [3 x i16] [i16 5, i16 123, i16 1024]
-@main.int16SliceSrc = internal unnamed_addr global { ptr, i64, i64 } { ptr @main.int16SliceSrc.buf, i64 3, i64 3 }
-@main.int16SliceDst = internal unnamed_addr global { ptr, i64, i64 } zeroinitializer
-@main.sliceSrcUntaint.buf = internal global [2 x i8] c"ab"
-@main.sliceDstUntaint.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcTaint.buf = internal global [2 x i8] c"cd"
-@main.sliceDstTaint.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal1.buf = external global [2 x i8]
-@main.sliceDstExternal1.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal2.buf = internal global [2 x i8] zeroinitializer
-@main.sliceDstExternal2.buf = external global [2 x i8]
-
-declare i64 @runtime.sliceCopy(ptr %dst, ptr %src, i64 %dstLen, i64 %srcLen, i64 %elemSize) unnamed_addr
-
-declare ptr @runtime.alloc(i64, ptr) unnamed_addr
-
-declare void @runtime.printuint8(i8)
-
-declare void @runtime.printint16(i16)
-
-declare void @use(ptr)
-
-define void @runtime.initAll() unnamed_addr {
-entry:
-  call void @main.init()
-  ret void
-}
-
-define void @main() unnamed_addr {
-entry:
-  ; print(uintSliceSrc[0])
-  %uint8SliceSrc.buf = load ptr, ptr @main.uint8SliceSrc
-  %uint8SliceSrc.val = load i8, ptr %uint8SliceSrc.buf
-  call void @runtime.printuint8(i8 %uint8SliceSrc.val)
-
-  ; print(uintSliceDst[0])
-  %uint8SliceDst.buf = load ptr, ptr @main.uint8SliceDst
-  %uint8SliceDst.val = load i8, ptr %uint8SliceDst.buf
-  call void @runtime.printuint8(i8 %uint8SliceDst.val)
-
-  ; print(int16SliceSrc[0])
-  %int16SliceSrc.buf = load ptr, ptr @main.int16SliceSrc
-  %int16SliceSrc.val = load i16, ptr %int16SliceSrc.buf
-  call void @runtime.printint16(i16 %int16SliceSrc.val)
-
-  ; print(int16SliceDst[0])
-  %int16SliceDst.buf = load ptr, ptr @main.int16SliceDst
-  %int16SliceDst.val = load i16, ptr %int16SliceDst.buf
-  call void @runtime.printint16(i16 %int16SliceDst.val)
-
-  ; print(sliceDstUntaint[0])
-  %sliceDstUntaint.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstUntaint.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstUntaint.val)
-
-  ; print(sliceDstTaint[0])
-  %sliceDstTaint.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstTaint.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstTaint.val)
-
-  ; print(sliceDstExternal1[0])
-  %sliceDstExternal1.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstExternal1.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstExternal1.val)
-
-  ; print(sliceDstExternal2[0])
-  %sliceDstExternal2.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstExternal2.buf, i32 0)
-  call void @runtime.printuint8(i8 %sliceDstExternal2.val)
-
-  ret void
-}
-
-define internal void @main.init() unnamed_addr {
-entry:
-  ; equivalent of:
-  ;     uint8SliceDst = make([]uint8, len(uint8SliceSrc))
-  %uint8SliceSrc = load { ptr, i64, i64 }, ptr @main.uint8SliceSrc
-  %uint8SliceSrc.len = extractvalue { ptr, i64, i64 } %uint8SliceSrc, 1
-  %uint8SliceDst.buf = call ptr @runtime.alloc(i64 %uint8SliceSrc.len, ptr null)
-  %0 = insertvalue { ptr, i64, i64 } undef, ptr %uint8SliceDst.buf, 0
-  %1 = insertvalue { ptr, i64, i64 } %0, i64 %uint8SliceSrc.len, 1
-  %2 = insertvalue { ptr, i64, i64 } %1, i64 %uint8SliceSrc.len, 2
-  store { ptr, i64, i64 } %2, ptr @main.uint8SliceDst
-
-  ; equivalent of:
-  ;     copy(uint8SliceDst, uint8SliceSrc)
-  %uint8SliceSrc.buf = extractvalue { ptr, i64, i64 } %uint8SliceSrc, 0
-  %copy.n = call i64 @runtime.sliceCopy(ptr %uint8SliceDst.buf, ptr %uint8SliceSrc.buf, i64 %uint8SliceSrc.len, i64 %uint8SliceSrc.len, i64 1)
-
-  ; equivalent of:
-  ;     int16SliceDst = make([]int16, len(int16SliceSrc))
-  %int16SliceSrc = load { ptr, i64, i64 }, ptr @main.int16SliceSrc
-  %int16SliceSrc.len = extractvalue { ptr, i64, i64 } %int16SliceSrc, 1
-  %int16SliceSrc.len.bytes = mul i64 %int16SliceSrc.len, 2
-  %int16SliceDst.buf = call ptr @runtime.alloc(i64 %int16SliceSrc.len.bytes, ptr null)
-  %3 = insertvalue { ptr, i64, i64 } undef, ptr %int16SliceDst.buf, 0
-  %4 = insertvalue { ptr, i64, i64 } %3, i64 %int16SliceSrc.len, 1
-  %5 = insertvalue { ptr, i64, i64 } %4, i64 %int16SliceSrc.len, 2
-  store { ptr, i64, i64 } %5, ptr @main.int16SliceDst
-
-  ; equivalent of:
-  ;     copy(int16SliceDst, int16SliceSrc)
-  %int16SliceSrc.buf = extractvalue { ptr, i64, i64 } %int16SliceSrc, 0
-  %copy.n2 = call i64 @runtime.sliceCopy(ptr %int16SliceDst.buf, ptr %int16SliceSrc.buf, i64 %int16SliceSrc.len, i64 %int16SliceSrc.len, i64 2)
-
-  ; Copy slice that has a known value.
-  %copy.n3 = call i64 @runtime.sliceCopy(ptr @main.sliceDstUntaint.buf, ptr @main.sliceSrcUntaint.buf, i64 2, i64 2, i64 1)
-
-  ; Copy slice that might have been modified by the external @use call.
-  ; This is a fix for https://github.com/tinygo-org/tinygo/issues/3890.
-  call void @use(ptr @main.sliceSrcTaint.buf)
-  %copy.n4 = call i64 @runtime.sliceCopy(ptr @main.sliceDstTaint.buf, ptr @main.sliceSrcTaint.buf, i64 2, i64 2, i64 1)
-
-  ; Test that copying from or into external buffers works correctly.
-  ; These copy operations must be done at runtime.
-  ; https://github.com/tinygo-org/tinygo/issues/4895
-  %copy.n5 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal1.buf, ptr @main.sliceSrcExternal1.buf, i64 2, i64 2, i64 1)
-  %copy.n6 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal2.buf, ptr @main.sliceSrcExternal2.buf, i64 2, i64 2, i64 1)
-
-  ret void
-}

interp/testdata/slice-copy.out.ll

@@ -1,42 +0,0 @@
-target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64--linux"
-
-@main.sliceSrcTaint.buf = internal global [2 x i8] c"cd"
-@main.sliceDstTaint.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal1.buf = external global [2 x i8]
-@main.sliceDstExternal1.buf = internal global [2 x i8] zeroinitializer
-@main.sliceSrcExternal2.buf = internal global [2 x i8] zeroinitializer
-@main.sliceDstExternal2.buf = external global [2 x i8]
-
-declare i64 @runtime.sliceCopy(ptr, ptr, i64, i64, i64) unnamed_addr
-
-declare void @runtime.printuint8(i8) local_unnamed_addr
-
-declare void @runtime.printint16(i16) local_unnamed_addr
-
-declare void @use(ptr) local_unnamed_addr
-
-define void @runtime.initAll() unnamed_addr {
-entry:
-  call void @use(ptr @main.sliceSrcTaint.buf)
-  %copy.n4 = call i64 @runtime.sliceCopy(ptr @main.sliceDstTaint.buf, ptr @main.sliceSrcTaint.buf, i64 2, i64 2, i64 1)
-  %copy.n5 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal1.buf, ptr @main.sliceSrcExternal1.buf, i64 2, i64 2, i64 1)
-  %copy.n6 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal2.buf, ptr @main.sliceSrcExternal2.buf, i64 2, i64 2, i64 1)
-  ret void
-}
-
-define void @main() unnamed_addr {
-entry:
-  call void @runtime.printuint8(i8 3)
-  call void @runtime.printuint8(i8 3)
-  call void @runtime.printint16(i16 5)
-  call void @runtime.printint16(i16 5)
-  call void @runtime.printuint8(i8 97)
-  %sliceDstTaint.val = load i8, ptr @main.sliceDstTaint.buf, align 1
-  call void @runtime.printuint8(i8 %sliceDstTaint.val)
-  %sliceDstExternal1.val = load i8, ptr @main.sliceDstExternal1.buf, align 1
-  call void @runtime.printuint8(i8 %sliceDstExternal1.val)
-  %sliceDstExternal2.val = load i8, ptr @main.sliceDstExternal2.buf, align 1
-  call void @runtime.printuint8(i8 %sliceDstExternal2.val)
-  ret void
-}

main.go

@@ -1635,6 +1635,7 @@ func main() {
 	cpuprofile := flag.String("cpuprofile", "", "cpuprofile output")
 	monitor := flag.Bool("monitor", false, "enable serial monitor")
 	baudrate := flag.Int("baudrate", 115200, "baudrate of serial monitor")
+	gocompatibility := flag.Bool("go-compatibility", true, "enable to check for Go versions compatibility, you can also configure this by setting the TINYGO_GOCOMPATIBILITY environment variable")
 
 	// Internal flags, that are only intended for TinyGo development.
 	printIR := flag.Bool("internal-printir", false, "print LLVM IR")
@@ -1712,6 +1713,16 @@ func main() {
 		ocdCommands = strings.Split(*ocdCommandsString, ",")
 	}
 
+	val, ok := os.LookupEnv("TINYGO_GOCOMPATIBILITY")
+	if ok {
+		b, err := strconv.ParseBool(val)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "could not parse TINYGO_GOCOMPATIBILITY value %q: %v\n", val, err)
+			os.Exit(1)
+		}
+		*gocompatibility = b
+	}
+
 	options := &compileopts.Options{
 		GOOS:            goenv.Get("GOOS"),
 		GOARCH:          goenv.Get("GOARCH"),
@@ -1748,6 +1759,7 @@ func main() {
 		Timeout:         *timeout,
 		WITPackage:      witPackage,
 		WITWorld:        witWorld,
+		GoCompatibility: *gocompatibility,
 	}
 	if *printCommands {
 		options.PrintCommands = printCommand

main_test.go

@@ -222,6 +222,7 @@ func TestBuild(t *testing.T) {
 			// to be sure.
 			t.Parallel()
 			options := optionsFromOSARCH("linux/mipsle/softfloat", sema)
+			emuCheck(t, options)
 			runTest("cgo/", options, t, nil, nil)
 		})
 	} else if runtime.GOOS == "windows" {

src/device/arm/arm.go

@@ -48,10 +48,10 @@ func Asm(asm string)
 // recognizes template values in the form {name}, like so:
 //
 //	arm.AsmFull(
-//	    "str {value}, {result}",
+//	    "str {value}, [{result}]",
 //	    map[string]interface{}{
-//	        "value":  1
-//	        "result": &dest,
+//	        "value":  1,
+//	        "result": uintptr(unsafe.Pointer(&dest)),
 //	    })
 //
 // You can use {} in the asm string (which expands to a register) to set the

src/device/arm/cortexm.S

@@ -1,29 +1,6 @@
 .syntax unified
 .cfi_sections .debug_frame
 
-.section .text.HardFault_Handler
-.global  HardFault_Handler
-.type    HardFault_Handler, %function
-HardFault_Handler:
-    .cfi_startproc
-    // Put the old stack pointer in the first argument, for easy debugging. This
-    // is especially useful on Cortex-M0, which supports far fewer debug
-    // facilities.
-    mov r0, sp
-
-    // Load the default stack pointer from address 0 so that we can call normal
-    // functions again that expect a working stack. However, it will corrupt the
-    // old stack so the function below must not attempt to recover from this
-    // fault.
-    movs r3, #0
-    ldr r3, [r3]
-    mov sp, r3
-
-    // Continue handling this error in Go.
-    bl handleHardFault
-    .cfi_endproc
-.size HardFault_Handler, .-HardFault_Handler
-
 // This is a convenience function for semihosting support.
 // At some point, this should be replaced by inline assembly.
 .section .text.SemihostingCall

src/device/arm64/arm64.go

@@ -10,10 +10,10 @@ func Asm(asm string)
 // recognizes template values in the form {name}, like so:
 //
 //	arm.AsmFull(
-//	    "str {value}, {result}",
+//	    "str {value}, [{result}]",
 //	    map[string]interface{}{
-//	        "value":  1
-//	        "result": &dest,
+//	        "value":  1,
+//	        "result": uintptr(unsafe.Pointer(&dest)),
 //	    })
 //
 // You can use {} in the asm string (which expands to a register) to set the

src/device/asm.go

@@ -10,10 +10,10 @@ func Asm(asm string)
 // recognizes template values in the form {name}, like so:
 //
 //	arm.AsmFull(
-//	    "str {value}, {result}",
+//	    "str {value}, [{result}]",
 //	    map[string]interface{}{
-//	        "value":  1
-//	        "result": &dest,
+//	        "value":  1,
+//	        "result": uintptr(unsafe.Pointer(&dest)),
 //	    })
 //
 // You can use {} in the asm string (which expands to a register) to set the

src/device/riscv/riscv.go

@@ -10,10 +10,10 @@ func Asm(asm string)
 // recognizes template values in the form {name}, like so:
 //
 //	arm.AsmFull(
-//	    "st {value}, {result}",
+//	    "st {value}, [{result}]",
 //	    map[string]interface{}{
-//	        "value":  1
-//	        "result": &dest,
+//	        "value":  1,
+//	        "result": uintptr(unsafe.Pointer(&dest)),
 //	    })
 //
 // You can use {} in the asm string (which expands to a register) to set the

src/examples/pwm/digispark.go

@@ -0,0 +1,12 @@
+//go:build digispark
+
+package main
+
+import "machine"
+
+var (
+	// Use Timer1 for PWM (recommended for ATtiny85)
+	pwm  = machine.Timer1
+	pinA = machine.P1 // PB1, Timer1 channel A (LED pin)
+	pinB = machine.P4 // PB4, Timer1 channel B
+)

src/internal/gclayout/gclayout.go

@@ -13,12 +13,14 @@ const (
 	// 64-bit int => bits = 6
 	sizeBits = 4 + unsafe.Sizeof(uintptr(0))/4
 
+	ptrAlign = unsafe.Alignof(uintptr(0))
+
 	sizeShift = sizeBits + 1
 
-	NoPtrs  = Layout(uintptr(0b0<<sizeShift) | uintptr(0b1<<1) | uintptr(1))
-	Pointer = Layout(uintptr(0b1<<sizeShift) | uintptr(0b1<<1) | uintptr(1))
-	String  = Layout(uintptr(0b01<<sizeShift) | uintptr(0b10<<1) | uintptr(1))
-	Slice   = Layout(uintptr(0b001<<sizeShift) | uintptr(0b11<<1) | uintptr(1))
+	NoPtrs  = Layout((0 << sizeShift) | (1 << 1) | 1)
+	Pointer = Layout((1 << sizeShift) | ((unsafe.Sizeof(unsafe.Pointer(nil)) / ptrAlign) << 1) | 1)
+	String  = Layout((1 << sizeShift) | ((unsafe.Sizeof("") / ptrAlign) << 1) | 1)
+	Slice   = Layout((1 << sizeShift) | ((unsafe.Sizeof([]byte{}) / ptrAlign) << 1) | 1)
 )
 
 func (l Layout) AsPtr() unsafe.Pointer { return unsafe.Pointer(l) }

src/internal/reflectlite/value.go

@@ -86,6 +86,64 @@ func (v Value) Interface() interface{} {
 	return valueInterfaceUnsafe(v)
 }
 
+func TypeAssert[T any](v Value) (T, bool) {
+	if v.typecode == nil {
+		panic("reflect.TypeAssert: zero Value")
+	}
+	if !v.isExported() {
+		// Do not allow access to unexported values via TypeAssert,
+		// because they might be pointers that should not be
+		// writable or methods or function that should not be callable.
+		panic("reflect.TypeAssert: cannot return value obtained from unexported field or method")
+	}
+
+	typ := TypeFor[T]()
+
+	// If v is an interface, return the element inside the interface.
+	//
+	// T is a concrete type and v is an interface. For example:
+	//
+	//	var v any = int(1)
+	//	val := ValueOf(&v).Elem()
+	//	TypeAssert[int](val) == val.Interface().(int)
+	//
+	// T is a interface and v is a non-nil interface value. For example:
+	//
+	//	var v any = &someError{}
+	//	val := ValueOf(&v).Elem()
+	//	TypeAssert[error](val) == val.Interface().(error)
+	//
+	// T is a interface and v is a nil interface value. For example:
+	//
+	//	var v error = nil
+	//	val := ValueOf(&v).Elem()
+	//	TypeAssert[error](val) == val.Interface().(error)
+	if v.Kind() == Interface {
+		val, ok := valueInterfaceUnsafe(v).(T)
+		return val, ok
+	}
+
+	// If T is an interface and v is a concrete type. For example:
+	//
+	//	TypeAssert[any](ValueOf(1)) == ValueOf(1).Interface().(any)
+	//	TypeAssert[error](ValueOf(&someError{})) == ValueOf(&someError{}).Interface().(error)
+	if typ.Kind() == Interface {
+		val, ok := valueInterfaceUnsafe(v).(T)
+		return val, ok
+	}
+
+	// Both v and T must be concrete types.
+	// The only way for an type-assertion to match is if the types are equal.
+	if typ != v.typecode {
+		var zero T
+		return zero, false
+	}
+	if !v.isIndirect() {
+		return *(*T)(unsafe.Pointer(&v.value)), true
+	}
+	return *(*T)(v.value), true
+}
+
 // valueInterfaceUnsafe is used by the runtime to hash map keys. It should not
 // be subject to the isExported check.
 func valueInterfaceUnsafe(v Value) interface{} {
@@ -1738,6 +1796,9 @@ func (e *ValueError) Error() string {
 //go:linkname memcpy runtime.memcpy
 func memcpy(dst, src unsafe.Pointer, size uintptr)
 
+//go:linkname memmove runtime.memmove
+func memmove(dst, src unsafe.Pointer, size uintptr)
+
 //go:linkname memzero runtime.memzero
 func memzero(ptr unsafe.Pointer, size uintptr)
 
@@ -1745,10 +1806,7 @@ func memzero(ptr unsafe.Pointer, size uintptr)
 func alloc(size uintptr, layout unsafe.Pointer) unsafe.Pointer
 
 //go:linkname sliceAppend runtime.sliceAppend
-func sliceAppend(srcBuf, elemsBuf unsafe.Pointer, srcLen, srcCap, elemsLen uintptr, elemSize uintptr) (unsafe.Pointer, uintptr, uintptr)
-
-//go:linkname sliceCopy runtime.sliceCopy
-func sliceCopy(dst, src unsafe.Pointer, dstLen, srcLen uintptr, elemSize uintptr) int
+func sliceAppend(srcBuf, elemsBuf unsafe.Pointer, srcLen, srcCap, elemsLen uintptr, elemSize uintptr, layout unsafe.Pointer) (unsafe.Pointer, uintptr, uintptr)
 
 // Copy copies the contents of src into dst until either
 // dst has been filled or src has been exhausted.
@@ -1779,7 +1837,10 @@ func Copy(dst, src Value) int {
 		dst.checkRO()
 	}
 
-	return sliceCopy(dstbuf, srcbuf, dstlen, srclen, dst.typecode.elem().Size())
+	minLen := min(dstlen, srclen)
+	elemSize := dst.typecode.elem().Size()
+	memmove(dstbuf, srcbuf, minLen*elemSize)
+	return int(minLen)
 }
 
 func buflen(v Value) (unsafe.Pointer, uintptr) {
@@ -1810,7 +1871,7 @@ func buflen(v Value) (unsafe.Pointer, uintptr) {
 }
 
 //go:linkname sliceGrow runtime.sliceGrow
-func sliceGrow(buf unsafe.Pointer, oldLen, oldCap, newCap, elemSize uintptr) (unsafe.Pointer, uintptr, uintptr)
+func sliceGrow(buf unsafe.Pointer, oldLen, oldCap, newCap, elemSize uintptr, layout unsafe.Pointer) (unsafe.Pointer, uintptr, uintptr)
 
 // extend slice to hold n new elements
 func extendSlice(v Value, n int) sliceHeader {
@@ -1823,7 +1884,10 @@ func extendSlice(v Value, n int) sliceHeader {
 		old = *(*sliceHeader)(v.value)
 	}
 
-	nbuf, nlen, ncap := sliceGrow(old.data, old.len, old.cap, old.len+uintptr(n), v.typecode.elem().Size())
+	elem := v.typecode.elem()
+	elemSize := elem.Size()
+	elemLayout := elem.gcLayout()
+	nbuf, nlen, ncap := sliceGrow(old.data, old.len, old.cap, old.len+uintptr(n), elemSize, elemLayout)
 
 	return sliceHeader{
 		data: nbuf,
@@ -1862,8 +1926,10 @@ func AppendSlice(s, t Value) Value {
 	}
 	sSlice := (*sliceHeader)(s.value)
 	tSlice := (*sliceHeader)(t.value)
-	elemSize := s.typecode.elem().Size()
-	ptr, len, cap := sliceAppend(sSlice.data, tSlice.data, sSlice.len, sSlice.cap, tSlice.len, elemSize)
+	elem := s.typecode.elem()
+	elemSize := elem.Size()
+	elemLayout := elem.gcLayout()
+	ptr, len, cap := sliceAppend(sSlice.data, tSlice.data, sSlice.len, sSlice.cap, tSlice.len, elemSize, elemLayout)
 	result := &sliceHeader{
 		data: ptr,
 		len:  len,

src/internal/task/task_asyncify.go

@@ -108,6 +108,10 @@ func (*stackState) unwind()
 func (t *Task) Resume() {
 	// The current task must be saved and restored because this can nest on WASM with JS.
 	prevTask := currentTask
+	if prevTask == nil {
+		// Save the system stack pointer.
+		saveStackPointer()
+	}
 	t.gcData.swap()
 	currentTask = t
 	if !t.state.launched {
@@ -123,6 +127,9 @@ func (t *Task) Resume() {
 	}
 }
 
+//go:linkname saveStackPointer runtime.saveStackPointer
+func saveStackPointer()
+
 //export tinygo_rewind
 func (*state) rewind()
 

src/internal/task/task_stack_esp32.go

@@ -1,4 +1,4 @@
-//go:build scheduler.tasks && esp32
+//go:build scheduler.tasks && (esp32 || esp32s3)
 
 package task
 

src/internal/task/task_threads.c

@@ -120,15 +120,21 @@ int tinygo_task_start(uintptr_t fn, void *args, void *task, pthread_t *thread, u
     #endif
     pthread_attr_t attrs;
     pthread_attr_init(&attrs);
+	pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
     pthread_attr_setstacksize(&attrs, stackSize);
     int result = pthread_create(thread, &attrs, &start_wrapper, &state);
     pthread_attr_destroy(&attrs);
+	if (result != 0) {
+		return result;
+	}
 
     // Wait until the thread has been created and read all state_pass variables.
     #if __APPLE__
     dispatch_semaphore_wait(state.startlock, DISPATCH_TIME_FOREVER);
+    dispatch_release(state.startlock);
     #else
     sem_wait(&state.startlock);
+    sem_destroy(&state.startlock);
     #endif
 
     return result;

src/internal/task/task_threads.go

@@ -23,16 +23,15 @@ type state struct {
 	// is needed to be able to scan the stack.
 	stackTop uintptr
 
+	// Lowest address of the stack.
+	// This is populated when the thread is stopped by the GC.
+	stackBottom uintptr
+
 	// Next task in the activeTasks queue.
 	QueueNext *Task
 
 	// Semaphore to pause/resume the thread atomically.
 	pauseSem Semaphore
-
-	// Semaphore used for stack scanning.
-	// We can't reuse pauseSem here since the thread might have been paused for
-	// other reasons (for example, because it was waiting on a channel).
-	gcSem Semaphore
 }
 
 // Goroutine counter, starting at 0 for the main goroutine.
@@ -96,6 +95,9 @@ func (t *Task) Resume() {
 	t.state.pauseSem.Post()
 }
 
+// otherGoroutines is the total number of live goroutines minus one.
+var otherGoroutines uint32
+
 // Start a new OS thread.
 func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	t := &Task{}
@@ -115,6 +117,7 @@ func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	}
 	t.state.QueueNext = activeTasks
 	activeTasks = t
+	otherGoroutines++
 	activeTaskLock.Unlock()
 }
 
@@ -135,6 +138,7 @@ func taskExited(t *Task) {
 			break
 		}
 	}
+	otherGoroutines--
 	activeTaskLock.Unlock()
 
 	// Sanity check.
@@ -143,9 +147,42 @@ func taskExited(t *Task) {
 	}
 }
 
-// Futex to wait on until all tasks have finished scanning the stack.
-// This is basically a sync.WaitGroup.
-var scanDoneFutex Futex
+// scanWaitGroup is used to wait on until all threads have finished the current state transition.
+var scanWaitGroup waitGroup
+
+type waitGroup struct {
+	f Futex
+}
+
+func initWaitGroup(n uint32) waitGroup {
+	var wg waitGroup
+	wg.f.Store(n)
+	return wg
+}
+
+func (wg *waitGroup) done() {
+	if wg.f.Add(^uint32(0)) == 0 {
+		wg.f.WakeAll()
+	}
+}
+
+func (wg *waitGroup) wait() {
+	for {
+		val := wg.f.Load()
+		if val == 0 {
+			return
+		}
+		wg.f.Wait(val)
+	}
+}
+
+// gcState is used to track and notify threads when the GC is stopping/resuming.
+var gcState Futex
+
+const (
+	gcStateResumed = iota
+	gcStateStopped
+)
 
 // GC scan phase. Because we need to stop the world while scanning, this kinda
 // needs to be done in the tasks package.
@@ -155,65 +192,71 @@ var scanDoneFutex Futex
 func GCStopWorldAndScan() {
 	current := Current()
 
-	// Don't allow new goroutines to be started while pausing/resuming threads
-	// in the stop-the-world phase.
-	activeTaskLock.Lock()
+	// NOTE: This does not need to be atomic.
+	if gcState.Load() == gcStateResumed {
+		// Don't allow new goroutines to be started while pausing/resuming threads
+		// in the stop-the-world phase.
+		activeTaskLock.Lock()
 
-	// Pause all other threads.
-	numOtherThreads := uint32(0)
-	for t := activeTasks; t != nil; t = t.state.QueueNext {
-		if t != current {
-			numOtherThreads++
-			tinygo_task_send_gc_signal(t.state.thread)
+		// Wait for threads to finish resuming.
+		scanWaitGroup.wait()
+
+		// Change the gc state to stopped.
+		// NOTE: This does not need to be atomic.
+		gcState.Store(gcStateStopped)
+
+		// Set the number of threads to wait for.
+		scanWaitGroup = initWaitGroup(otherGoroutines)
+
+		// Pause all other threads.
+		for t := activeTasks; t != nil; t = t.state.QueueNext {
+			if t != current {
+				tinygo_task_send_gc_signal(t.state.thread)
+			}
 		}
+
+		// Wait for the threads to finish stopping.
+		scanWaitGroup.wait()
 	}
 
-	// Store the number of threads to wait for in the futex.
-	// This is the equivalent of doing an initial wg.Add(numOtherThreads).
-	scanDoneFutex.Store(numOtherThreads)
+	// Scan other thread stacks.
+	for t := activeTasks; t != nil; t = t.state.QueueNext {
+		if t != current {
+			markRoots(t.state.stackBottom, t.state.stackTop)
+		}
+	}
 
 	// Scan the current stack, and all current registers.
 	scanCurrentStack()
 
-	// Wake each paused thread for the first time so it will scan the stack.
-	for t := activeTasks; t != nil; t = t.state.QueueNext {
-		if t != current {
-			t.state.gcSem.Post()
-		}
-	}
-
-	// Wait until all threads have finished scanning their stack.
-	// This is the equivalent of wg.Wait()
-	for {
-		val := scanDoneFutex.Load()
-		if val == 0 {
-			break
-		}
-		scanDoneFutex.Wait(val)
-	}
-
 	// Scan all globals (implemented in the runtime).
 	gcScanGlobals()
 }
 
 // After the GC is done scanning, resume all other threads.
-//
-// This must only be called after a GCStopWorldAndScan call.
 func GCResumeWorld() {
-	current := Current()
-
-	// Wake each paused thread for the second time, so they will resume normal
-	// operation.
-	for t := activeTasks; t != nil; t = t.state.QueueNext {
-		if t != current {
-			t.state.gcSem.Post()
-		}
+	// NOTE: This does not need to be atomic.
+	if gcState.Load() == gcStateResumed {
+		// This is already resumed.
+		return
 	}
 
+	// Set the wait group to track resume progress.
+	scanWaitGroup = initWaitGroup(otherGoroutines)
+
+	// Set the state to resumed.
+	gcState.Store(gcStateResumed)
+
+	// Wake all of the stopped threads.
+	gcState.WakeAll()
+
 	// Allow goroutines to start and exit again.
 	activeTaskLock.Unlock()
 }
 
+//go:linkname markRoots runtime.markRoots
+func markRoots(start, end uintptr)
+
 // Scan globals, implemented in the runtime package.
 func gcScanGlobals()
 
@@ -221,34 +264,27 @@ var stackScanLock PMutex
 
 //export tinygo_task_gc_pause
 func tingyo_task_gc_pause(sig int32) {
-	// Wait until we get the signal to start scanning the stack.
-	Current().state.gcSem.Wait()
+	// Write the entrty stack pointer to the state.
+	Current().state.stackBottom = uintptr(stacksave())
 
-	// Scan the thread stack.
-	// Only scan a single thread stack at a time, because the GC marking phase
-	// doesn't support parallelism.
-	// TODO: it may be possible to call markRoots directly (without saving
-	// registers) since we are in a signal handler that already saved a bunch of
-	// registers. This is an optimization left for a future time.
-	stackScanLock.Lock()
-	scanCurrentStack()
-	stackScanLock.Unlock()
+	// Notify the GC that we are stopped.
+	scanWaitGroup.done()
 
-	// Equivalent of wg.Done(): subtract one from the futex and if the result is
-	// 0 (meaning we were the last in the waitgroup), wake the waiting thread.
-	n := uint32(1)
-	if scanDoneFutex.Add(-n) == 0 {
-		scanDoneFutex.Wake()
+	// Wait for the GC to resume.
+	for gcState.Load() == gcStateStopped {
+		gcState.Wait(gcStateStopped)
 	}
 
-	// Wait until we get the signal we can resume normally (after the mark phase
-	// has finished).
-	Current().state.gcSem.Wait()
+	// Notify the GC that we have resumed.
+	scanWaitGroup.done()
 }
 
 //go:export tinygo_scanCurrentStack
 func scanCurrentStack()
 
+//go:linkname stacksave runtime.stacksave
+func stacksave() unsafe.Pointer
+
 // Return the highest address of the current stack.
 func StackTop() uintptr {
 	return Current().state.stackTop

src/machine/board_amken_trio.go

@@ -0,0 +1,153 @@
+//go:build amken_trio
+
+// RabbitPNP Toolhead Board
+// MCU: STM32G0B1CBTx (LQFP48, 128KB Flash, 144KB RAM)
+
+package machine
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+)
+
+// Vacuum sensors (PWM input via TIM2)
+const (
+	VAC1 = PA0 // TIM2_CH1
+	VAC2 = PA1 // TIM2_CH2
+)
+
+// Motor 1 pins (stepper driver)
+const (
+	M1_CS   = PC7
+	M1_DIR  = PB13 // REFL
+	M1_STEP = PB14 // REFR
+	M1_ENN  = PA10 // Enable (active low)
+)
+
+// Motor 2 pins (stepper driver)
+const (
+	M2_CS   = PA9
+	M2_DIR  = PB12 // REFL
+	M2_STEP = PB11 // REFR
+	M2_ENN  = PC6  // Enable (active low)
+)
+
+// Motor 3 pins (stepper driver)
+const (
+	M3_CS   = PB15
+	M3_DIR  = PB2 // REFL
+	M3_STEP = PB1 // REFR
+	M3_ENN  = PA8 // Enable (active low)
+)
+
+// LED
+const (
+	LED         = LED1
+	LED_BUILTIN = LED1
+	LED1        = PB7
+)
+
+// Solenoid and Neopixel (PWM via TIM4)
+const (
+	SOLENOID = PB8 // TIM4_CH3
+	NEOPIXEL = PB9 // TIM4_CH4
+)
+
+// Endstops
+const (
+	ENDSTOP_IN1 = PC13
+	ENDSTOP_IN2 = PC14
+)
+
+// Magnetic sensor
+const (
+	MAG1 = PB3
+)
+
+// Accelerometer chip select (LIS2D on SPI2)
+const (
+	LIS2D_CS = PB5
+)
+
+// SPI1 pins (motor drivers)
+const (
+	SPI1_SCK_PIN = PA5
+	SPI1_SDO_PIN = PA2 // MOSI
+	SPI1_SDI_PIN = PA6 // MISO
+	SPI0_SCK_PIN = SPI1_SCK_PIN
+	SPI0_SDO_PIN = SPI1_SDO_PIN
+	SPI0_SDI_PIN = SPI1_SDI_PIN
+)
+
+// SPI2 pins (accelerometer)
+const (
+	SPI2_SCK_PIN = PB10
+	SPI2_SDO_PIN = PA4 // MOSI
+	SPI2_SDI_PIN = PA3 // MISO
+)
+
+// I2C2 pins
+const (
+	I2C2_SCL_PIN = PA7
+	I2C2_SDA_PIN = PB4
+	I2C0_SCL_PIN = I2C2_SCL_PIN
+	I2C0_SDA_PIN = I2C2_SDA_PIN
+)
+
+// FDCAN1 pins
+const (
+	CAN_RX = PD0
+	CAN_TX = PD1
+)
+
+// USB pins
+const (
+	USB_DM = PA11
+	USB_DP = PA12
+)
+
+// UART pins (not directly connected but required by machine package)
+const (
+	UART_TX_PIN = NoPin
+	UART_RX_PIN = NoPin
+)
+
+var (
+	// SPI1 for motor drivers
+	SPI1 = &SPI{
+		Bus:             stm32.SPI1,
+		AltFuncSelector: AF0_SYSTEM,
+	}
+	SPI0 = SPI1
+
+	// SPI2 for accelerometer
+	SPI2 = &SPI{
+		Bus:             stm32.SPI2,
+		AltFuncSelector: AF1_TIM1_TIM2_TIM3_LPTIM1,
+	}
+
+	// I2C2
+	I2C2 = &I2C{
+		Bus:             stm32.I2C2,
+		AltFuncSelector: AF6_SPI2_USART3_USART4_I2C1,
+	}
+	I2C0 = I2C2
+
+	// FDCAN1 on PD0 (RX) / PD1 (TX) with onboard transceiver
+	CAN1  = &_CAN1
+	_CAN1 = CAN{
+		Bus:             stm32.FDCAN1,
+		TxAltFuncSelect: AF3_FDCAN1_FDCAN2,
+		RxAltFuncSelect: AF3_FDCAN1_FDCAN2,
+		instance:        0,
+	}
+	// Alias for convenience
+	CAN0 = CAN1
+)
+
+// Suppress unused import warning for interrupt package
+var _ = interrupt.New
+
+func init() {
+	// No UART configured on this board - uses USB or CAN for communication
+}

src/machine/board_digispark.go

@@ -2,17 +2,26 @@
 
 package machine
 
+// Digispark is a tiny ATtiny85-based board with 6 I/O pins.
+//
+// PWM is available on the following pins:
+//   - P0 (PB0): Timer0 channel A
+//   - P1 (PB1): Timer0 channel B or Timer1 channel A (LED pin)
+//   - P4 (PB4): Timer1 channel B
+//
+// Timer1 is recommended for PWM as it provides more flexible frequency control.
+
 // Return the current CPU frequency in hertz.
 func CPUFrequency() uint32 {
 	return 16000000
 }
 
 const (
-	P0 Pin = PB0
-	P1 Pin = PB1
+	P0 Pin = PB0 // PWM available (Timer0 OC0A)
+	P1 Pin = PB1 // PWM available (Timer0 OC0B or Timer1 OC1A)
 	P2 Pin = PB2
 	P3 Pin = PB3
-	P4 Pin = PB4
+	P4 Pin = PB4 // PWM available (Timer1 OC1B)
 	P5 Pin = PB5
 
 	LED = P1

src/machine/board_esp32s3-wroom1.go

@@ -0,0 +1,15 @@
+//go:build esp32s3_wroom1
+
+package machine
+
+const (
+	SPI1_SCK_PIN  = GPIO12 // SCK
+	SPI1_MOSI_PIN = GPIO11 // SDO (MOSI)
+	SPI1_MISO_PIN = GPIO13 // SDI (MISO)
+	SPI1_CS_PIN   = GPIO10 // CS
+
+	SPI2_SCK_PIN  = GPIO36 // SCK
+	SPI2_MOSI_PIN = GPIO35 // SDO (MOSI)
+	SPI2_MISO_PIN = GPIO37 // SDI (MISO)
+	SPI2_CS_PIN   = GPIO34 // CS
+)

src/machine/board_fe310.go

@@ -15,8 +15,8 @@ const (
 	P09 Pin = 9
 	P10 Pin = 10
 	P11 Pin = 11
-	P12 Pin = 12
-	P13 Pin = 13
+	P12 Pin = 12 // peripherals: I2C0 SDA
+	P13 Pin = 13 // peripherals: I2C0 SCL
 	P14 Pin = 14
 	P15 Pin = 15
 	P16 Pin = 16

src/machine/board_feather-m0.go

@@ -43,6 +43,15 @@ const (
 	USBCDC_DP_PIN = PA25
 )
 
+// UART0 pins
+const (
+	UART0_TX_PIN = D1
+	UART0_RX_PIN = D0
+)
+
+// UART0 on the Feather M0.
+var UART0 = &sercomUSART0
+
 // UART1 pins
 const (
 	UART_TX_PIN = D10

src/machine/board_nucleog0b1re.go

@@ -0,0 +1,131 @@
+//go:build nucleog0b1re
+
+// Schematic: https://www.st.com/resource/en/user_manual/um2324-stm32-nucleo64-boards-mb1360-stmicroelectronics.pdf
+// Datasheet: https://www.st.com/resource/en/datasheet/stm32g0b1re.pdf
+
+package machine
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+)
+
+const (
+	// Arduino Pins
+	A0 = PA0
+	A1 = PA1
+	A2 = PA4
+	A3 = PB1
+	A4 = PA11
+	A5 = PA12
+
+	D0  = PB7
+	D1  = PB6
+	D2  = PA10
+	D3  = PB3
+	D4  = PB5
+	D5  = PB4
+	D6  = PB10
+	D7  = PA8
+	D8  = PA9
+	D9  = PC7
+	D10 = PB0
+	D11 = PA7
+	D12 = PA6
+	D13 = PA5
+	D14 = PB9
+	D15 = PB8
+)
+
+// User LD4: the green LED is a user LED connected to ARDUINO signal D13 corresponding
+// to STM32 I/O PA5.
+const (
+	LED         = LED_BUILTIN
+	LED_BUILTIN = LED_GREEN
+	LED_GREEN   = PA5
+)
+
+// User B1: the user button is connected to PC13.
+const (
+	BUTTON = PC13
+)
+
+const (
+	// UART pins
+	// PA2 and PA3 are connected to the ST-Link Virtual Com Port (VCP)
+	UART_TX_PIN = PA2
+	UART_RX_PIN = PA3
+
+	// I2C pins
+	// PB8 is SCL (connected to Arduino connector D15)
+	// PB9 is SDA (connected to Arduino connector D14)
+	I2C0_SCL_PIN = PB8
+	I2C0_SDA_PIN = PB9
+
+	// SPI pins
+	SPI1_SCK_PIN = PA5
+	SPI1_SDI_PIN = PA6
+	SPI1_SDO_PIN = PA7
+	SPI0_SCK_PIN = SPI1_SCK_PIN
+	SPI0_SDI_PIN = SPI1_SDI_PIN
+	SPI0_SDO_PIN = SPI1_SDO_PIN
+
+	// CAN pins (directly accessible on Nucleo-G0B1RE board)
+	// FDCAN1: PA11 (TX) / PA12 (RX) using AF9
+	// FDCAN2: PD12 (TX) / PD13 (RX) using AF3
+	CAN1_TX_PIN = PA11
+	CAN1_RX_PIN = PA12
+	CAN2_TX_PIN = PD12
+	CAN2_RX_PIN = PD13
+)
+
+var (
+	// USART2 is the hardware serial port connected to the onboard ST-LINK
+	// debugger to be exposed as virtual COM port over USB on Nucleo boards.
+	UART1  = &_UART1
+	_UART1 = UART{
+		Buffer:            NewRingBuffer(),
+		Bus:               stm32.USART2,
+		TxAltFuncSelector: AF1_TIM1_TIM2_TIM3_LPTIM1,
+		RxAltFuncSelector: AF1_TIM1_TIM2_TIM3_LPTIM1,
+	}
+	DefaultUART = UART1
+
+	// I2C1 is documented, alias to I2C0 as well
+	I2C1 = &I2C{
+		Bus:             stm32.I2C1,
+		AltFuncSelector: AF6_SPI2_USART3_USART4_I2C1,
+	}
+	I2C0 = I2C1
+
+	// SPI1 is documented, alias to SPI0 as well
+	SPI1 = &SPI{
+		Bus:             stm32.SPI1,
+		AltFuncSelector: AF0_SYSTEM,
+	}
+	SPI0 = SPI1
+
+	// FDCAN1 on PA11 (TX) / PA12 (RX)
+	CAN1  = &_CAN1
+	_CAN1 = CAN{
+		Bus:             stm32.FDCAN1,
+		TxAltFuncSelect: AF9_FDCAN1_FDCAN2,
+		RxAltFuncSelect: AF9_FDCAN1_FDCAN2,
+		instance:        0,
+	}
+
+	// FDCAN2 on PD12 (TX) / PD13 (RX)
+	CAN2  = &_CAN2
+	_CAN2 = CAN{
+		Bus:             stm32.FDCAN2,
+		TxAltFuncSelect: AF3_FDCAN1_FDCAN2,
+		RxAltFuncSelect: AF3_FDCAN1_FDCAN2,
+		instance:        1,
+	}
+)
+
+func init() {
+	UART1.Interrupt = interrupt.New(stm32.IRQ_USART2_LPUART2, _UART1.handleInterrupt)
+	// Note: FDCAN interrupts share with USB (IRQ_UCPD1_UCPD2_USB = 8)
+	// User should configure interrupts via SetInterrupt method if needed
+}

src/machine/board_pico2_ice.go

@@ -0,0 +1,173 @@
+//go:build pico2_ice
+
+// Most of the info is from
+// https://pico2-ice.tinyvision.ai/md_pinout.html although
+// (2025-09-07) RP4 appears twice in that pinout - the schematic is
+// more clear. Consistent with other RPi boards, we use GPn instead of
+// RPn to reference the RPi connected pins.
+
+package machine
+
+// GPIO pins
+const (
+	GP0  = GPIO0
+	GP1  = GPIO1
+	GP2  = GPIO2
+	GP3  = GPIO3
+	GP4  = GPIO4
+	GP5  = GPIO5
+	GP6  = GPIO6
+	GP7  = GPIO7
+	GP8  = GPIO8
+	GP9  = GPIO9
+	GP10 = GPIO10
+	GP11 = GPIO11
+	GP12 = GPIO12
+	GP13 = GPIO13
+	GP14 = GPIO14
+	GP15 = GPIO15
+	GP16 = GPIO16
+	GP17 = GPIO17
+	GP18 = GPIO18
+	GP19 = GPIO19
+	GP20 = GPIO20
+	GP21 = GPIO21
+	GP22 = GPIO22
+	GP23 = GPIO23
+	GP24 = GPIO24
+	GP25 = GPIO25
+	GP26 = GPIO26
+	GP27 = GPIO27
+	GP28 = GPIO28
+	GP29 = GPIO29
+	GP30 = GPIO30
+	GP31 = GPIO31
+	GP32 = GPIO32
+	GP33 = GPIO33
+	GP34 = GPIO34
+	GP35 = GPIO35
+	GP36 = GPIO36
+	GP37 = GPIO37
+	GP38 = GPIO38
+	GP39 = GPIO39
+	GP40 = GPIO40
+	GP41 = GPIO41
+	GP42 = GPIO42
+	GP43 = GPIO43
+	GP44 = GPIO44
+	GP45 = GPIO45
+	GP46 = GPIO46
+	GP47 = GPIO47
+
+	// RPi pins shared with ICE. The ICE number is what appears on
+	// the board silkscreen.
+	ICE9  = GP28
+	ICE11 = GP29
+	ICE14 = GP7
+	ICE15 = GP6
+	ICE16 = GP5
+	ICE17 = GP4
+	ICE18 = GP27
+	ICE19 = GP23
+	ICE20 = GP22
+	ICE21 = GP26
+	ICE23 = GP25
+	ICE25 = GP30
+	ICE26 = GP24
+	ICE27 = GP20
+
+	// FPGA Clock pin.
+	ICE35_G0 = GP21
+
+	// Silkscreen & Pinout names
+	ICE_SSN = ICE16
+	ICE_SO  = ICE14
+	ICE_SI  = ICE17
+	ICE_CK  = ICE15
+	SD      = GP2
+	SC      = GP3
+
+	FPGA_RSTN = GP31
+	A3        = GP32
+	A1        = GP33
+	A4        = GP34
+	A2        = GP35
+	B3        = GP36
+	B1        = GP37
+	B4        = GP38
+	B2        = GP39
+	N0        = GP40 // On the board these are labeled "~0"
+	N1        = GP41
+	N2        = GP42
+	N3        = GP43
+	N4        = GP44
+	N5        = GP45
+	N6        = GP46
+
+	// Functions from Schematic.
+	ICE_DONE = GP40
+	USB_BOOT = GP42
+
+	// Button
+	SW1     = GP42
+	BOOTSEL = GP42
+
+	// Tricolor LEDs
+	LED_RED   = GP1
+	LED_GREEN = GP0
+	LED_BLUE  = GP9
+
+	// Onboard LED
+	LED = LED_GREEN
+
+	// Onboard crystal oscillator frequency, in MHz.
+	xoscFreq = 12 // MHz
+)
+
+// This board does not define default i2c pins.
+const (
+	I2C0_SDA_PIN = NoPin
+	I2C0_SCL_PIN = NoPin
+	I2C1_SDA_PIN = NoPin
+	I2C1_SCL_PIN = NoPin
+)
+
+// SPI default pins
+const (
+	// Default Serial Clock Bus 0 for SPI communications
+	SPI0_SCK_PIN = GPIO18
+	// Default Serial Out Bus 0 for SPI communications
+	SPI0_SDO_PIN = GPIO19 // Tx
+	// Default Serial In Bus 0 for SPI communications
+	SPI0_SDI_PIN = GPIO16 // Rx
+
+	// Default Serial Clock Bus 1 for SPI communications
+	SPI1_SCK_PIN = GPIO10
+	// Default Serial Out Bus 1 for SPI communications
+	SPI1_SDO_PIN = GPIO11 // Tx
+	// Default Serial In Bus 1 for SPI communications
+	SPI1_SDI_PIN = GPIO12 // Rx
+)
+
+// UART pins
+const (
+	UART0_TX_PIN = GPIO0
+	UART0_RX_PIN = GPIO1
+	UART1_TX_PIN = GPIO8
+	UART1_RX_PIN = GPIO9
+	UART_TX_PIN  = UART0_TX_PIN
+	UART_RX_PIN  = UART0_RX_PIN
+)
+
+var DefaultUART = UART0
+
+// USB identifiers
+const (
+	usb_STRING_PRODUCT      = "Pico2"
+	usb_STRING_MANUFACTURER = "Raspberry Pi"
+)
+
+var (
+	usb_VID uint16 = 0x2E8A
+	usb_PID uint16 = 0x000A
+)

src/machine/board_vicharak_shrike-lite.go

@@ -0,0 +1,118 @@
+//go:build vicharak_shrike_lite
+
+// Pin mappings for Vicharak Shrike-Lite.
+//
+// Reference: https://vicharak-in.github.io/shrike/shrike_pinouts.html
+
+package machine
+
+// Digital
+const (
+	IO0  Pin = GPIO0
+	IO1  Pin = GPIO1
+	IO2  Pin = GPIO2
+	IO3  Pin = GPIO3
+	IO4  Pin = GPIO4
+	IO5  Pin = GPIO5
+	IO6  Pin = GPIO6
+	IO7  Pin = GPIO7
+	IO8  Pin = GPIO8
+	IO9  Pin = GPIO9
+	IO10 Pin = GPIO10
+	IO11 Pin = GPIO11
+	IO12 Pin = GPIO12
+	IO13 Pin = GPIO13
+	IO14 Pin = GPIO14
+	IO15 Pin = GPIO15
+	IO16 Pin = GPIO16
+	IO17 Pin = GPIO17
+	IO18 Pin = GPIO18
+	IO19 Pin = GPIO19
+	IO20 Pin = GPIO20
+	IO21 Pin = GPIO21
+	IO22 Pin = GPIO22
+	IO23 Pin = GPIO23
+	IO24 Pin = GPIO24
+	IO25 Pin = GPIO25
+	IO26 Pin = GPIO26
+	IO27 Pin = GPIO27
+	IO28 Pin = GPIO28
+	IO29 Pin = GPIO29
+)
+
+// FPGA Pins
+const (
+	FPGA_EN  Pin = IO13
+	FPGA_PWR Pin = IO12
+	// SPI_SCLK
+	F3 Pin = IO2
+	// SPI_SS
+	F4 Pin = IO1
+	// SPI_SI (MOSI)
+	F5 Pin = IO3
+	// SPI_SO (MISO) / CONFIG
+	F6  Pin = IO0
+	F18 Pin = IO14
+	F17 Pin = IO15
+)
+
+// Analog pins
+const (
+	A0 Pin = IO26
+	A1 Pin = IO27
+	A2 Pin = IO28
+	A3 Pin = IO29
+)
+
+// LED
+const (
+	LED = IO4
+)
+
+// I2C pins
+const (
+	I2C0_SDA_PIN Pin = IO24
+	I2C0_SCL_PIN Pin = IO25
+
+	I2C1_SDA_PIN Pin = IO6
+	I2C1_SCL_PIN Pin = IO7
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN Pin = IO18
+	SPI0_SDO_PIN Pin = IO19
+	SPI0_SDI_PIN Pin = IO20
+
+	SPI1_SCK_PIN Pin = IO10
+	SPI1_SDO_PIN Pin = IO11
+	SPI1_SDI_PIN Pin = IO8
+)
+
+// Onboard crystal oscillator frequency, in MHz.
+const (
+	xoscFreq = 12 // MHz
+)
+
+// UART pins
+const (
+	UART0_TX_PIN = IO28
+	UART0_RX_PIN = IO29
+	UART_TX_PIN  = UART0_TX_PIN
+	UART_RX_PIN  = UART0_RX_PIN
+	UART1_TX_PIN = IO24
+	UART1_RX_PIN = IO25
+)
+
+var DefaultUART = UART0
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "Shrike-Lite"
+	usb_STRING_MANUFACTURER = "Vicharak"
+)
+
+var (
+	usb_VID uint16 = 0x2e8a
+	usb_PID uint16 = 0x0003
+)

src/machine/board_xiao-esp32s3.go

@@ -0,0 +1,64 @@
+//go:build xiao_esp32s3
+
+// This file contains the pin mappings for the Seeed XIAO ESP32S3 boards.
+//
+// Seeed Studio XIAO ESP32S3 is an IoT mini development board based on
+// the Espressif ESP32-S3 WiFi/Bluetooth dual-mode chip.
+//
+// - https://www.seeedstudio.com/XIAO-ESP32S3-p-5627.html
+// - https://wiki.seeedstudio.com/xiao_esp32s3_getting_started/
+
+package machine
+
+// Digital Pins
+const (
+	D0  = GPIO1
+	D1  = GPIO2
+	D2  = GPIO3
+	D3  = GPIO4
+	D4  = GPIO5
+	D5  = GPIO6
+	D6  = GPIO43
+	D7  = GPIO44
+	D8  = GPIO7
+	D9  = GPIO8
+	D10 = GPIO9
+)
+
+// Analog pins
+const (
+	A0 = GPIO1
+	A1 = GPIO2
+	A2 = GPIO3
+	A3 = GPIO4
+)
+
+// UART pins
+const (
+	UART_RX_PIN = GPIO44
+	UART_TX_PIN = GPIO43
+)
+
+// I2C pins
+const (
+	SDA_PIN = GPIO5
+	SCL_PIN = GPIO6
+)
+
+// SPI pins
+const (
+	SPI1_SCK_PIN  = GPIO7 // D8
+	SPI1_MISO_PIN = GPIO8 // D9
+	SPI1_MOSI_PIN = GPIO9 // D10
+	SPI1_CS_PIN   = NoPin
+
+	SPI2_SCK_PIN  = NoPin
+	SPI2_MOSI_PIN = NoPin
+	SPI2_MISO_PIN = NoPin
+	SPI2_CS_PIN   = NoPin
+)
+
+// Onboard LEDs
+const (
+	LED = GPIO21
+)

src/machine/can.go

@@ -0,0 +1,102 @@
+//go:build stm32g0
+
+package machine
+
+// unexported functions here are implemented in the device file
+// and added to the build tags of this file.
+
+// These types are an alias for documentation purposes exclusively. We wish
+// the interface to be used by other ecosystems besides TinyGo which is why
+// we need these types to be a primitive types at the interface level.
+// If these types are defined at machine or machine/can level they are not
+// usable by non-TinyGo projects. This is not good news for fostering wider adoption
+// of our API in "big-Go" embedded system projects like TamaGo and periph.io
+type (
+	// CAN IDs in tinygo are represented as 30 bit integers where
+	// bits 1..29 store the actual ID and the 30th bit stores the IDE bit (if extended ID).
+	// We include the extended ID bit in the ID itself to make comparison of IDs easier for users
+	// since two identical IDs where one is extended and one is not are NOT equivalent IDs.
+	canID = uint32
+	// CAN flags bitmask are defined below.
+	canFlags = uint32
+)
+
+// CAN ID definitions.
+const (
+	canIDStdMask      canID = (1 << 11) - 1
+	canIDExtendedMask canID = (1 << 29) - 1
+	canIDExtendedBit  canID = 1 << 30
+)
+
+// CAN Flag bit definitions.
+const (
+	canFlagBRS canFlags = 1 << 0 // Bit Rate Switch active on tx/rx of frame.
+	canFlagFDF canFlags = 1 << 1 // Is a FD Frame.
+	canFlagRTR canFlags = 1 << 2 // is a retransmission request frame.
+	canFlagESI canFlags = 1 << 3 // Error status indicator active on tx/rx of frame.
+	canFlagIDE canFlags = 1 << 4 // Extended ID.
+)
+
+// TxFIFOLevel returns amount of CAN frames stored for transmission and total Tx fifo length.
+func (can *CAN) TxFIFOLevel() (level int, maxlevel int) {
+	return can.txFIFOLevel()
+}
+
+// Tx puts a CAN frame in TxFIFO for transmission. Returns error if TxFIFO is full.
+func (can *CAN) Tx(id canID, flags canFlags, data []byte) error {
+	return can.tx(id, flags, data)
+}
+
+// RxFIFOLevel returns amount of CAN frames received and stored and total Rx fifo length.
+// If the hardware is interrupt driven RxFIFOLevel should return 0,0.
+func (can *CAN) RxFIFOLevel() (level int, maxlevel int) {
+	return can.rxFIFOLevel()
+}
+
+type canRxCallback = func(data []byte, id canID, timestamp uint32, flags canFlags)
+
+// SetRxCallback sets the receive callback. See [canFlags] for information on how bits are layed out.
+func (can *CAN) SetRxCallback(cb canRxCallback) {
+	can.setRxCallback(cb)
+}
+
+// RxPoll is called periodically for poll driven drivers. If the driver is interrupt driven
+// then RxPoll is a no-op and may return nil. Users may determine if a CAN is interrupt driven by
+// checking if RxFIFOLevel returns 0,0.
+func (can *CAN) RxPoll() error {
+	return can.rxPoll()
+}
+
+// DLC to bytes lookup table
+var dlcToBytes = [16]byte{0, 1, 2, 3, 4, 5, 6, 7, 8, 12, 16, 20, 24, 32, 48, 64}
+
+// dlcToLength converts a DLC value to actual byte length
+func dlcToLength(dlc byte) uint8 {
+	if dlc > 15 {
+		dlc = 15
+	}
+	return dlcToBytes[dlc]
+}
+
+// lengthToDLC converts a byte length to DLC value
+func lengthToDLC(length uint8) (dlc byte) {
+	switch {
+	case length <= 8:
+		dlc = length
+	case length <= 12:
+		dlc = 9
+	case length <= 16:
+		dlc = 10
+	case length <= 20:
+		dlc = 11
+	case length <= 24:
+		dlc = 12
+	case length <= 32:
+		dlc = 13
+	case length <= 48:
+		dlc = 14
+	default:
+		dlc = 15
+	}
+	return dlc
+}

src/machine/flash.go

@@ -1,4 +1,4 @@
-//go:build nrf || nrf51 || nrf52 || nrf528xx || stm32f4 || stm32l4 || stm32wlx || atsamd21 || atsamd51 || atsame5x || rp2040 || rp2350
+//go:build nrf || nrf51 || nrf52 || nrf528xx || stm32f4 || stm32l0 || stm32l4 || stm32wlx || atsamd21 || atsamd51 || atsame5x || rp2040 || rp2350
 
 package machine
 
@@ -43,6 +43,11 @@ type BlockDevice interface {
 	io.ReaderAt
 
 	// WriteAt writes the given number of bytes to the block device.
+	//
+	// This interface directly writes data to the underlying block device.
+	// Different kinds of devices have different requirements: most can only
+	// write data after the page has been erased, and many can only write data
+	// with specific alignment (such as 4-byte alignment).
 	io.WriterAt
 
 	// Size returns the number of bytes in this block device.

src/machine/machine.go

@@ -1,6 +1,9 @@
 package machine
 
-import "errors"
+import (
+	"errors"
+	"unsafe"
+)
 
 var (
 	ErrTimeoutRNG         = errors.New("machine: RNG Timeout")
@@ -62,3 +65,30 @@ func (p Pin) Low() {
 type ADC struct {
 	Pin Pin
 }
+
+// Convert the pointer to a uintptr, to be used for memory I/O (DMA for
+// example). It also means the pointer is "gone" as far as the compiler is
+// concerned, and a GC cycle might deallocate the object. To prevent this from
+// happening, also call keepAliveNoEscape at a point after the address isn't
+// accessed anymore by the hardware.
+// The only exception is if the pointer is accessed later in a volatile way
+// (volatile read/write), which also forces the value to stay alive until that
+// point.
+//
+// This function is treated specially by the compiler to mark the 'ptr'
+// parameter as not escaping.
+//
+// TODO: this function should eventually be replaced with the proposed ptrtoaddr
+// instruction in LLVM. See:
+// https://discourse.llvm.org/t/clarifiying-the-semantics-of-ptrtoint/83987/10
+// https://github.com/llvm/llvm-project/pull/139357
+func unsafeNoEscape(ptr unsafe.Pointer) uintptr {
+	return uintptr(ptr)
+}
+
+// Make sure the given pointer stays alive until this point. This is similar to
+// runtime.KeepAlive, with the difference that it won't let the pointer escape.
+// This is typically used together with unsafeNoEscape.
+//
+// This is a compiler intrinsic.
+func keepAliveNoEscape(ptr unsafe.Pointer)

src/machine/machine_atsamd21_usb.go

@@ -337,7 +337,7 @@ func handleUSBSetAddress(setup usb.Setup) bool {
 
 // SendUSBInPacket sends a packet for USB (interrupt in / bulk in).
 func SendUSBInPacket(ep uint32, data []byte) bool {
-	sendUSBPacket(ep, data, 0)
+	sendUSBPacket(ep, data)
 
 	// clear transfer complete flag
 	setEPINTFLAG(ep, sam.USB_DEVICE_EPINTFLAG_TRCPT1)
@@ -351,27 +351,28 @@ func SendUSBInPacket(ep uint32, data []byte) bool {
 // Prevent file size increases: https://github.com/tinygo-org/tinygo/pull/998
 //
 //go:noinline
-func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
-	l := uint16(len(data))
-	if 0 < maxsize && maxsize < l {
-		l = maxsize
+func sendUSBPacket(ep uint32, data []byte) {
+	// Select the corresponding buffer.
+	buffer := udd_ep_control_cache_buffer[:]
+	if ep != 0 {
+		buffer = udd_ep_in_cache_buffer[ep][:]
 	}
 
-	// Set endpoint address for sending data
-	if ep == 0 {
-		copy(udd_ep_control_cache_buffer[:], data[:l])
-		usbEndpointDescriptors[ep].DeviceDescBank[1].ADDR.Set(uint32(uintptr(unsafe.Pointer(&udd_ep_control_cache_buffer))))
-	} else {
-		copy(udd_ep_in_cache_buffer[ep][:], data[:l])
-		usbEndpointDescriptors[ep].DeviceDescBank[1].ADDR.Set(uint32(uintptr(unsafe.Pointer(&udd_ep_in_cache_buffer[ep]))))
-	}
+	// Copy the packet to the buffer.
+	copy(buffer[:len(data)], data)
+
+	// Select the corresponding endpoint descriptor.
+	endpoint := &usbEndpointDescriptors[ep].DeviceDescBank[1]
+
+	// Set the endpoint address.
+	endpoint.ADDR.Set(uint32(uintptr(unsafe.Pointer(unsafe.SliceData(buffer)))))
 
 	// clear multi-packet size which is total bytes already sent
-	usbEndpointDescriptors[ep].DeviceDescBank[1].PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Mask << usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Pos)
+	endpoint.PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Mask << usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Pos)
 
 	// set byte count, which is total number of bytes to be sent
-	usbEndpointDescriptors[ep].DeviceDescBank[1].PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
-	usbEndpointDescriptors[ep].DeviceDescBank[1].PCKSIZE.SetBits((uint32(l) & usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask) << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
+	endpoint.PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
+	endpoint.PCKSIZE.SetBits((uint32(len(data)) & usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask) << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
 }
 
 func ReceiveUSBControlPacket() ([cdcLineInfoSize]byte, error) {

src/machine/machine_atsamd51_usb.go

@@ -340,7 +340,7 @@ func handleUSBSetAddress(setup usb.Setup) bool {
 
 // SendUSBInPacket sends a packet for USB (interrupt in / bulk in).
 func SendUSBInPacket(ep uint32, data []byte) bool {
-	sendUSBPacket(ep, data, 0)
+	sendUSBPacket(ep, data)
 
 	// clear transfer complete flag
 	setEPINTFLAG(ep, sam.USB_DEVICE_ENDPOINT_EPINTFLAG_TRCPT1)
@@ -354,27 +354,28 @@ func SendUSBInPacket(ep uint32, data []byte) bool {
 // Prevent file size increases: https://github.com/tinygo-org/tinygo/pull/998
 //
 //go:noinline
-func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
-	l := uint16(len(data))
-	if 0 < maxsize && maxsize < l {
-		l = maxsize
+func sendUSBPacket(ep uint32, data []byte) {
+	// Select the corresponding buffer.
+	buffer := udd_ep_control_cache_buffer[:]
+	if ep != 0 {
+		buffer = udd_ep_in_cache_buffer[ep][:]
 	}
 
-	// Set endpoint address for sending data
-	if ep == 0 {
-		copy(udd_ep_control_cache_buffer[:], data[:l])
-		usbEndpointDescriptors[ep].DeviceDescBank[1].ADDR.Set(uint32(uintptr(unsafe.Pointer(&udd_ep_control_cache_buffer))))
-	} else {
-		copy(udd_ep_in_cache_buffer[ep][:], data[:l])
-		usbEndpointDescriptors[ep].DeviceDescBank[1].ADDR.Set(uint32(uintptr(unsafe.Pointer(&udd_ep_in_cache_buffer[ep]))))
-	}
+	// Copy the packet to the buffer.
+	copy(buffer[:len(data)], data)
+
+	// Select the corresponding endpoint descriptor.
+	endpoint := &usbEndpointDescriptors[ep].DeviceDescBank[1]
+
+	// Set the endpoint address.
+	endpoint.ADDR.Set(uint32(uintptr(unsafe.Pointer(unsafe.SliceData(buffer)))))
 
 	// clear multi-packet size which is total bytes already sent
-	usbEndpointDescriptors[ep].DeviceDescBank[1].PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Mask << usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Pos)
+	endpoint.PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Mask << usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Pos)
 
 	// set byte count, which is total number of bytes to be sent
-	usbEndpointDescriptors[ep].DeviceDescBank[1].PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
-	usbEndpointDescriptors[ep].DeviceDescBank[1].PCKSIZE.SetBits((uint32(l) & usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask) << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
+	endpoint.PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
+	endpoint.PCKSIZE.SetBits((uint32(len(data)) & usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask) << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
 }
 
 func ReceiveUSBControlPacket() ([cdcLineInfoSize]byte, error) {

src/machine/machine_attiny85.go

@@ -21,3 +21,524 @@ func (p Pin) getPortMask() (*volatile.Register8, uint8) {
 	// Very simple for the attiny85, which only has a single port.
 	return avr.PORTB, 1 << uint8(p)
 }
+
+// PWM is one PWM peripheral, which consists of a counter and two output
+// channels (that can be connected to two fixed pins). You can set the frequency
+// using SetPeriod, but only for all the channels in this PWM peripheral at
+// once.
+type PWM struct {
+	num uint8
+}
+
+var (
+	Timer0 = PWM{0} // 8 bit timer for PB0 and PB1
+	Timer1 = PWM{1} // 8 bit high-speed timer for PB1 and PB4
+)
+
+// GTCCR bits for Timer1 that are not defined in the device file
+const (
+	gtccrPWM1B  = 0x40 // Pulse Width Modulator B Enable
+	gtccrCOM1B0 = 0x10 // Comparator B Output Mode bit 0
+	gtccrCOM1B1 = 0x20 // Comparator B Output Mode bit 1
+)
+
+// Configure enables and configures this PWM.
+//
+// For Timer0, there is only a limited number of periods available, namely the
+// CPU frequency divided by 256 and again divided by 1, 8, 64, 256, or 1024.
+// For a MCU running at 8MHz, this would be a period of 32µs, 256µs, 2048µs,
+// 8192µs, or 32768µs.
+//
+// For Timer1, the period is more flexible as it uses OCR1C as the top value.
+// Timer1 also supports more prescaler values (1 to 16384).
+func (pwm PWM) Configure(config PWMConfig) error {
+	switch pwm.num {
+	case 0: // Timer/Counter 0 (8-bit)
+		// Calculate the timer prescaler.
+		var prescaler uint8
+		switch config.Period {
+		case 0, (uint64(1e9) * 256 * 1) / uint64(CPUFrequency()):
+			prescaler = 1
+		case (uint64(1e9) * 256 * 8) / uint64(CPUFrequency()):
+			prescaler = 2
+		case (uint64(1e9) * 256 * 64) / uint64(CPUFrequency()):
+			prescaler = 3
+		case (uint64(1e9) * 256 * 256) / uint64(CPUFrequency()):
+			prescaler = 4
+		case (uint64(1e9) * 256 * 1024) / uint64(CPUFrequency()):
+			prescaler = 5
+		default:
+			return ErrPWMPeriodTooLong
+		}
+
+		avr.TCCR0B.Set(prescaler)
+		// Set the PWM mode to fast PWM (mode = 3).
+		avr.TCCR0A.Set(avr.TCCR0A_WGM00 | avr.TCCR0A_WGM01)
+
+	case 1: // Timer/Counter 1 (8-bit high-speed)
+		// Timer1 on ATtiny85 is different from ATmega328:
+		// - It's 8-bit with configurable top (OCR1C)
+		// - Has more prescaler options (1-16384)
+		// - PWM mode is enabled per-channel via PWM1A/PWM1B bits
+		var top uint64
+		if config.Period == 0 {
+			// Use a top appropriate for LEDs.
+			top = 0xff
+		} else {
+			// Calculate top value: top = period * (CPUFrequency / 1e9)
+			top = config.Period * (uint64(CPUFrequency()) / 1000000) / 1000
+		}
+
+		// Timer1 prescaler values: 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384
+		const maxTop = 256
+		var prescaler uint8
+		switch {
+		case top <= maxTop:
+			prescaler = 1 // prescaler 1
+		case top/2 <= maxTop:
+			prescaler = 2 // prescaler 2
+			top /= 2
+		case top/4 <= maxTop:
+			prescaler = 3 // prescaler 4
+			top /= 4
+		case top/8 <= maxTop:
+			prescaler = 4 // prescaler 8
+			top /= 8
+		case top/16 <= maxTop:
+			prescaler = 5 // prescaler 16
+			top /= 16
+		case top/32 <= maxTop:
+			prescaler = 6 // prescaler 32
+			top /= 32
+		case top/64 <= maxTop:
+			prescaler = 7 // prescaler 64
+			top /= 64
+		case top/128 <= maxTop:
+			prescaler = 8 // prescaler 128
+			top /= 128
+		case top/256 <= maxTop:
+			prescaler = 9 // prescaler 256
+			top /= 256
+		case top/512 <= maxTop:
+			prescaler = 10 // prescaler 512
+			top /= 512
+		case top/1024 <= maxTop:
+			prescaler = 11 // prescaler 1024
+			top /= 1024
+		case top/2048 <= maxTop:
+			prescaler = 12 // prescaler 2048
+			top /= 2048
+		case top/4096 <= maxTop:
+			prescaler = 13 // prescaler 4096
+			top /= 4096
+		case top/8192 <= maxTop:
+			prescaler = 14 // prescaler 8192
+			top /= 8192
+		case top/16384 <= maxTop:
+			prescaler = 15 // prescaler 16384
+			top /= 16384
+		default:
+			return ErrPWMPeriodTooLong
+		}
+
+		// Set prescaler (CS1[3:0] bits)
+		avr.TCCR1.Set(prescaler)
+		// Set top value
+		avr.OCR1C.Set(uint8(top - 1))
+	}
+	return nil
+}
+
+// SetPeriod updates the period of this PWM peripheral.
+// To set a particular frequency, use the following formula:
+//
+//	period = 1e9 / frequency
+//
+// If you use a period of 0, a period that works well for LEDs will be picked.
+//
+// SetPeriod will not change the prescaler, but also won't change the current
+// value in any of the channels. This means that you may need to update the
+// value for the particular channel.
+//
+// Note that you cannot pick any arbitrary period after the PWM peripheral has
+// been configured. If you want to switch between frequencies, pick the lowest
+// frequency (longest period) once when calling Configure and adjust the
+// frequency here as needed.
+func (pwm PWM) SetPeriod(period uint64) error {
+	if pwm.num == 0 {
+		return ErrPWMPeriodTooLong // Timer0 doesn't support dynamic period
+	}
+
+	// Timer1 can adjust period via OCR1C
+	var top uint64
+	if period == 0 {
+		top = 0xff
+	} else {
+		top = period * (uint64(CPUFrequency()) / 1000000) / 1000
+	}
+
+	// Get current prescaler
+	prescaler := avr.TCCR1.Get() & 0x0f
+	// Timer1 prescaler values follow a power-of-2 pattern:
+	// prescaler n maps to divisor 2^(n-1), so we can use a simple shift
+	if prescaler > 0 && prescaler <= 15 {
+		top >>= (prescaler - 1)
+	}
+
+	if top > 256 {
+		return ErrPWMPeriodTooLong
+	}
+
+	avr.OCR1C.Set(uint8(top - 1))
+	avr.TCNT1.Set(0)
+
+	return nil
+}
+
+// Top returns the current counter top, for use in duty cycle calculation. It
+// will only change with a call to Configure or SetPeriod, otherwise it is
+// constant.
+//
+// The value returned here is hardware dependent. In general, it's best to treat
+// it as an opaque value that can be divided by some number and passed to Set
+// (see Set documentation for more information).
+func (pwm PWM) Top() uint32 {
+	if pwm.num == 1 {
+		// Timer1 has configurable top via OCR1C
+		return uint32(avr.OCR1C.Get()) + 1
+	}
+	// Timer0 goes from 0 to 0xff (256 in total)
+	return 256
+}
+
+// Counter returns the current counter value of the timer in this PWM
+// peripheral. It may be useful for debugging.
+func (pwm PWM) Counter() uint32 {
+	switch pwm.num {
+	case 0:
+		return uint32(avr.TCNT0.Get())
+	case 1:
+		return uint32(avr.TCNT1.Get())
+	}
+	return 0
+}
+
+// Prescaler lookup tables using uint16 (more efficient than uint64 on AVR)
+// Timer0 prescaler lookup table (index 0-7 maps to prescaler bits)
+var timer0Prescalers = [8]uint16{0, 1, 8, 64, 256, 1024, 0, 0}
+
+// Timer1 prescaler lookup table (index 0-15 maps to prescaler bits)
+var timer1Prescalers = [16]uint16{0, 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384}
+
+// Period returns the used PWM period in nanoseconds. It might deviate slightly
+// from the configured period due to rounding.
+func (pwm PWM) Period() uint64 {
+	var prescaler uint64
+	switch pwm.num {
+	case 0:
+		prescalerBits := avr.TCCR0B.Get() & 0x7
+		prescaler = uint64(timer0Prescalers[prescalerBits])
+		if prescaler == 0 {
+			return 0
+		}
+	case 1:
+		prescalerBits := avr.TCCR1.Get() & 0x0f
+		prescaler = uint64(timer1Prescalers[prescalerBits])
+		if prescaler == 0 {
+			return 0
+		}
+	}
+	top := uint64(pwm.Top())
+	return prescaler * top * 1000 / uint64(CPUFrequency()/1e6)
+}
+
+// Channel returns a PWM channel for the given pin.
+func (pwm PWM) Channel(pin Pin) (uint8, error) {
+	pin.Configure(PinConfig{Mode: PinOutput})
+	pin.Low()
+	switch pwm.num {
+	case 0:
+		switch pin {
+		case PB0: // OC0A
+			avr.TCCR0A.SetBits(avr.TCCR0A_COM0A1)
+			return 0, nil
+		case PB1: // OC0B
+			avr.TCCR0A.SetBits(avr.TCCR0A_COM0B1)
+			return 1, nil
+		}
+	case 1:
+		switch pin {
+		case PB1: // OC1A
+			// Enable PWM on channel A
+			avr.TCCR1.SetBits(avr.TCCR1_PWM1A | avr.TCCR1_COM1A1)
+			return 0, nil
+		case PB4: // OC1B
+			// Enable PWM on channel B (controlled via GTCCR)
+			avr.GTCCR.SetBits(gtccrPWM1B | gtccrCOM1B1)
+			return 1, nil
+		}
+	}
+	return 0, ErrInvalidOutputPin
+}
+
+// SetInverting sets whether to invert the output of this channel.
+// Without inverting, a 25% duty cycle would mean the output is high for 25% of
+// the time and low for the rest. Inverting flips the output as if a NOT gate
+// was placed at the output, meaning that the output would be 25% low and 75%
+// high with a duty cycle of 25%.
+func (pwm PWM) SetInverting(channel uint8, inverting bool) {
+	switch pwm.num {
+	case 0:
+		switch channel {
+		case 0: // channel A, PB0
+			if inverting {
+				avr.PORTB.SetBits(1 << 0)
+				avr.TCCR0A.SetBits(avr.TCCR0A_COM0A0)
+			} else {
+				avr.PORTB.ClearBits(1 << 0)
+				avr.TCCR0A.ClearBits(avr.TCCR0A_COM0A0)
+			}
+		case 1: // channel B, PB1
+			if inverting {
+				avr.PORTB.SetBits(1 << 1)
+				avr.TCCR0A.SetBits(avr.TCCR0A_COM0B0)
+			} else {
+				avr.PORTB.ClearBits(1 << 1)
+				avr.TCCR0A.ClearBits(avr.TCCR0A_COM0B0)
+			}
+		}
+	case 1:
+		switch channel {
+		case 0: // channel A, PB1
+			if inverting {
+				avr.PORTB.SetBits(1 << 1)
+				avr.TCCR1.SetBits(avr.TCCR1_COM1A0)
+			} else {
+				avr.PORTB.ClearBits(1 << 1)
+				avr.TCCR1.ClearBits(avr.TCCR1_COM1A0)
+			}
+		case 1: // channel B, PB4
+			if inverting {
+				avr.PORTB.SetBits(1 << 4)
+				avr.GTCCR.SetBits(gtccrCOM1B0)
+			} else {
+				avr.PORTB.ClearBits(1 << 4)
+				avr.GTCCR.ClearBits(gtccrCOM1B0)
+			}
+		}
+	}
+}
+
+// Set updates the channel value. This is used to control the channel duty
+// cycle, in other words the fraction of time the channel output is high (or low
+// when inverted). For example, to set it to a 25% duty cycle, use:
+//
+//	pwm.Set(channel, pwm.Top() / 4)
+//
+// pwm.Set(channel, 0) will set the output to low and pwm.Set(channel,
+// pwm.Top()) will set the output to high, assuming the output isn't inverted.
+func (pwm PWM) Set(channel uint8, value uint32) {
+	switch pwm.num {
+	case 0:
+		switch channel {
+		case 0: // channel A, PB0
+			if value == 0 {
+				avr.TCCR0A.ClearBits(avr.TCCR0A_COM0A1)
+			} else {
+				avr.OCR0A.Set(uint8(value - 1))
+				avr.TCCR0A.SetBits(avr.TCCR0A_COM0A1)
+			}
+		case 1: // channel B, PB1
+			if value == 0 {
+				avr.TCCR0A.ClearBits(avr.TCCR0A_COM0B1)
+			} else {
+				avr.OCR0B.Set(uint8(value - 1))
+				avr.TCCR0A.SetBits(avr.TCCR0A_COM0B1)
+			}
+		}
+	case 1:
+		switch channel {
+		case 0: // channel A, PB1
+			if value == 0 {
+				avr.TCCR1.ClearBits(avr.TCCR1_COM1A1)
+			} else {
+				avr.OCR1A.Set(uint8(value - 1))
+				avr.TCCR1.SetBits(avr.TCCR1_COM1A1)
+			}
+		case 1: // channel B, PB4
+			if value == 0 {
+				avr.GTCCR.ClearBits(gtccrCOM1B1)
+			} else {
+				avr.OCR1B.Set(uint8(value - 1))
+				avr.GTCCR.SetBits(gtccrCOM1B1)
+			}
+		}
+	}
+}
+
+// SPIConfig is used to store config info for SPI.
+type SPIConfig struct {
+	Frequency uint32
+	LSBFirst  bool
+	Mode      uint8
+}
+
+// SPI is the USI-based SPI implementation for ATTiny85.
+// The ATTiny85 doesn't have dedicated SPI hardware, but uses the USI
+// (Universal Serial Interface) in three-wire mode.
+//
+// Fixed pin mapping (directly controlled by USI hardware):
+//   - PB2: SCK (clock)
+//   - PB1: DO/MOSI (data out)
+//   - PB0: DI/MISO (data in)
+//
+// Note: CS pin must be managed by the user.
+type SPI struct {
+	// Delay cycles for frequency control (0 = max speed)
+	delayCycles uint16
+
+	// USICR value configured for the selected SPI mode
+	usicrValue uint8
+
+	// LSB-first mode (requires software bit reversal)
+	lsbFirst bool
+}
+
+// SPI0 is the USI-based SPI interface on the ATTiny85
+var SPI0 = SPI{}
+
+// Configure sets up the USI for SPI communication.
+// Note: The user must configure and control the CS pin separately.
+func (s *SPI) Configure(config SPIConfig) error {
+	// Configure USI pins (fixed by hardware)
+	// PB1 (DO/MOSI) -> OUTPUT
+	// PB2 (USCK/SCK) -> OUTPUT
+	// PB0 (DI/MISO) -> INPUT
+	PB1.Configure(PinConfig{Mode: PinOutput})
+	PB2.Configure(PinConfig{Mode: PinOutput})
+	PB0.Configure(PinConfig{Mode: PinInput})
+
+	// Reset USI registers
+	avr.USIDR.Set(0)
+	avr.USISR.Set(0)
+
+	// Configure USI for SPI mode:
+	// - USIWM0: Three-wire mode (SPI)
+	// - USICS1: External clock source (software controlled via USITC)
+	// - USICLK: Clock strobe - enables counter increment on USITC toggle
+	// - USICS0: Controls clock phase (CPHA)
+	//
+	// SPI Modes:
+	//   Mode 0 (CPOL=0, CPHA=0): Clock idle low, sample on rising edge
+	//   Mode 1 (CPOL=0, CPHA=1): Clock idle low, sample on falling edge
+	//   Mode 2 (CPOL=1, CPHA=0): Clock idle high, sample on falling edge
+	//   Mode 3 (CPOL=1, CPHA=1): Clock idle high, sample on rising edge
+	//
+	// For USI, USICS0 controls the sampling edge when USICS1=1:
+	//   USICS0=0: Positive edge (rising)
+	//   USICS0=1: Negative edge (falling)
+	switch config.Mode {
+	case Mode0: // CPOL=0, CPHA=0: idle low, sample rising
+		PB2.Low()
+		s.usicrValue = avr.USICR_USIWM0 | avr.USICR_USICS1 | avr.USICR_USICLK
+	case Mode1: // CPOL=0, CPHA=1: idle low, sample falling
+		PB2.Low()
+		s.usicrValue = avr.USICR_USIWM0 | avr.USICR_USICS1 | avr.USICR_USICS0 | avr.USICR_USICLK
+	case Mode2: // CPOL=1, CPHA=0: idle high, sample falling
+		PB2.High()
+		s.usicrValue = avr.USICR_USIWM0 | avr.USICR_USICS1 | avr.USICR_USICS0 | avr.USICR_USICLK
+	case Mode3: // CPOL=1, CPHA=1: idle high, sample rising
+		PB2.High()
+		s.usicrValue = avr.USICR_USIWM0 | avr.USICR_USICS1 | avr.USICR_USICLK
+	default: // Default to Mode 0
+		PB2.Low()
+		s.usicrValue = avr.USICR_USIWM0 | avr.USICR_USICS1 | avr.USICR_USICLK
+	}
+	avr.USICR.Set(s.usicrValue)
+
+	// Calculate delay cycles for frequency control
+	// Each bit transfer requires 2 clock toggles (rising + falling edge)
+	// The loop overhead is approximately 10-15 cycles per toggle on AVR
+	// We calculate additional delay cycles needed to achieve the target frequency
+	if config.Frequency > 0 && config.Frequency < CPUFrequency()/2 {
+		// Cycles per half-period = CPUFrequency / (2 * Frequency)
+		// Subtract loop overhead (~15 cycles) to get delay cycles
+		cyclesPerHalfPeriod := CPUFrequency() / (2 * config.Frequency)
+		const loopOverhead = 15
+		if cyclesPerHalfPeriod > loopOverhead {
+			s.delayCycles = uint16(cyclesPerHalfPeriod - loopOverhead)
+		} else {
+			s.delayCycles = 0
+		}
+	} else {
+		// Max speed - no delay
+		s.delayCycles = 0
+	}
+
+	// Store LSBFirst setting for use in Transfer
+	s.lsbFirst = config.LSBFirst
+
+	return nil
+}
+
+// reverseByte reverses the bit order of a byte (MSB <-> LSB)
+// Used for LSB-first SPI mode since USI hardware only supports MSB-first
+func reverseByte(b byte) byte {
+	b = (b&0xF0)>>4 | (b&0x0F)<<4
+	b = (b&0xCC)>>2 | (b&0x33)<<2
+	b = (b&0xAA)>>1 | (b&0x55)<<1
+	return b
+}
+
+// Transfer performs a single byte SPI transfer (send and receive simultaneously)
+// This implements the USI-based SPI transfer using the "clock strobing" technique
+func (s *SPI) Transfer(b byte) (byte, error) {
+	// For LSB-first mode, reverse the bits before sending
+	// USI hardware only supports MSB-first, so we do it in software
+	if s.lsbFirst {
+		b = reverseByte(b)
+	}
+
+	// Load the byte to transmit into the USI Data Register
+	avr.USIDR.Set(b)
+
+	// Clear the counter overflow flag by writing 1 to it (AVR quirk)
+	// This also resets the 4-bit counter to 0
+	avr.USISR.Set(avr.USISR_USIOIF)
+
+	// Clock the data out/in
+	// We need 16 clock toggles (8 bits × 2 edges per bit)
+	// The USI counter counts each clock edge, so it overflows at 16
+	// After 16 toggles, the clock returns to its idle state (set by CPOL in Configure)
+	//
+	// IMPORTANT: Only toggle USITC here!
+	// - USITC toggles the clock pin
+	// - The USICR mode bits (USIWM0, USICS1, USICS0, USICLK) were set in Configure()
+	// - SetBits preserves those bits and only sets USITC
+	if s.delayCycles == 0 {
+		// Fast path: no delay, run at maximum speed
+		for !avr.USISR.HasBits(avr.USISR_USIOIF) {
+			avr.USICR.SetBits(avr.USICR_USITC)
+		}
+	} else {
+		// Frequency-controlled path: add delay between clock toggles
+		for !avr.USISR.HasBits(avr.USISR_USIOIF) {
+			avr.USICR.SetBits(avr.USICR_USITC)
+			// Delay loop for frequency control
+			// Each iteration is approximately 3 cycles on AVR (dec, brne)
+			for i := s.delayCycles; i > 0; i-- {
+				avr.Asm("nop")
+			}
+		}
+	}
+
+	// Get the received byte
+	result := avr.USIDR.Get()
+
+	// For LSB-first mode, reverse the received bits
+	if s.lsbFirst {
+		result = reverseByte(result)
+	}
+
+	return result, nil
+}

src/machine/machine_esp32c3.go

@@ -509,98 +509,6 @@ func (uart *UART) writeByte(b byte) error {
 
 func (uart *UART) flush() {}
 
-type Serialer interface {
-	WriteByte(c byte) error
-	Write(data []byte) (n int, err error)
-	Configure(config UARTConfig) error
-	Buffered() int
-	ReadByte() (byte, error)
-	DTR() bool
-	RTS() bool
-}
-
-// USB Serial/JTAG Controller
-// See esp32-c3_technical_reference_manual_en.pdf
-// pg. 736
-type USB_DEVICE struct {
-	Bus *esp.USB_DEVICE_Type
-}
-
-var (
-	_USBCDC = &USB_DEVICE{
-		Bus: esp.USB_DEVICE,
-	}
-
-	USBCDC Serialer = _USBCDC
-)
-
-var (
-	errUSBWrongSize            = errors.New("USB: invalid write size")
-	errUSBCouldNotWriteAllData = errors.New("USB: could not write all data")
-	errUSBBufferEmpty          = errors.New("USB: read buffer empty")
-)
-
-func (usbdev *USB_DEVICE) Configure(config UARTConfig) error {
-	return nil
-}
-
-func (usbdev *USB_DEVICE) WriteByte(c byte) error {
-	if usbdev.Bus.GetEP1_CONF_SERIAL_IN_EP_DATA_FREE() == 0 {
-		return errUSBCouldNotWriteAllData
-	}
-
-	usbdev.Bus.SetEP1_RDWR_BYTE(uint32(c))
-	usbdev.flush()
-
-	return nil
-}
-
-func (usbdev *USB_DEVICE) Write(data []byte) (n int, err error) {
-	if len(data) == 0 || len(data) > 64 {
-		return 0, errUSBWrongSize
-	}
-
-	for i, c := range data {
-		if usbdev.Bus.GetEP1_CONF_SERIAL_IN_EP_DATA_FREE() == 0 {
-			if i > 0 {
-				usbdev.flush()
-			}
-
-			return i, errUSBCouldNotWriteAllData
-		}
-		usbdev.Bus.SetEP1_RDWR_BYTE(uint32(c))
-	}
-
-	usbdev.flush()
-	return len(data), nil
-}
-
-func (usbdev *USB_DEVICE) Buffered() int {
-	return int(usbdev.Bus.GetEP1_CONF_SERIAL_OUT_EP_DATA_AVAIL())
-}
-
-func (usbdev *USB_DEVICE) ReadByte() (byte, error) {
-	if usbdev.Bus.GetEP1_CONF_SERIAL_OUT_EP_DATA_AVAIL() != 0 {
-		return byte(usbdev.Bus.GetEP1_RDWR_BYTE()), nil
-	}
-
-	return 0, nil
-}
-
-func (usbdev *USB_DEVICE) DTR() bool {
-	return false
-}
-
-func (usbdev *USB_DEVICE) RTS() bool {
-	return false
-}
-
-func (usbdev *USB_DEVICE) flush() {
-	usbdev.Bus.SetEP1_CONF_WR_DONE(1)
-	for usbdev.Bus.GetEP1_CONF_SERIAL_IN_EP_DATA_FREE() == 0 {
-	}
-}
-
 // GetRNG returns 32-bit random numbers using the ESP32-C3 true random number generator,
 // Random numbers are generated based on the thermal noise in the system and the
 // asynchronous clock mismatch.

src/machine/machine_esp32s3.go

@@ -0,0 +1,310 @@
+//go:build esp32s3
+
+package machine
+
+import (
+	"device/esp"
+	"errors"
+	"runtime/volatile"
+	"unsafe"
+)
+
+const deviceName = esp.Device
+
+const xtalClock = 40_000000       // 40MHz
+const apbClock = 80_000000        // 80MHz
+const cryptoPWMClock = 160_000000 // 160MHz
+
+// GetCPUFrequency returns the current CPU frequency of the chip.
+func GetCPUFrequency() (uint32, error) {
+	switch esp.SYSTEM.GetSYSCLK_CONF_SOC_CLK_SEL() {
+	case 0:
+		return xtalClock / (esp.SYSTEM.GetSYSCLK_CONF_PRE_DIV_CNT() + 1), nil
+	case 1:
+		switch esp.SYSTEM.GetCPU_PER_CONF_CPUPERIOD_SEL() {
+		case 0:
+			return 80e6, nil
+		case 1:
+			return 160e6, nil
+		case 2:
+			// If esp.SYSTEM.GetCPU_PER_CONF_PLL_FREQ_SEL() == 1, this is undefined
+			return 240e6, nil
+		}
+	case 2:
+		//RC Fast Clock
+		return (175e5) / (esp.SYSTEM.GetSYSCLK_CONF_PRE_DIV_CNT() + 1), nil
+	}
+	return 0, errors.New("machine: Unable to determine current cpu frequency")
+}
+
+// SetCPUFrequency sets the frequency of the CPU to one of several targets
+func SetCPUFrequency(frequency uint32) error {
+	// Always assume we are on PLL. Lower frequencies can be set with a different
+	// clock source, but this will change the behavior of APB clock and Crypto PWM
+	// clock
+	//esp.SYSTEM.SetSYSCLK_CONF_SOC_CLK_SEL(1)
+
+	switch frequency {
+	case 80_000000:
+		esp.SYSTEM.SetCPU_PER_CONF_CPUPERIOD_SEL(0)
+		esp.SYSTEM.SetCPU_PER_CONF_PLL_FREQ_SEL(0) // Reduce PLL freq when possible
+		return nil
+	case 160_000000:
+		esp.SYSTEM.SetCPU_PER_CONF_CPUPERIOD_SEL(1)
+		esp.SYSTEM.SetCPU_PER_CONF_PLL_FREQ_SEL(0)
+		return nil
+	case 240_000000:
+		esp.SYSTEM.SetCPU_PER_CONF_PLL_FREQ_SEL(1) // Increase PLL freq when needed
+		esp.SYSTEM.SetCPU_PER_CONF_CPUPERIOD_SEL(2)
+		return nil
+	}
+	return errors.New("machine: Unsupported CPU frequency selected. Supported: 80, 160, 240 MHz")
+}
+
+var (
+	ErrInvalidSPIBus = errors.New("machine: invalid SPI bus")
+)
+
+const (
+	PinOutput PinMode = iota
+	PinInput
+	PinInputPullup
+	PinInputPulldown
+)
+
+// Hardware pin numbers
+const (
+	GPIO0  Pin = 0
+	GPIO1  Pin = 1
+	GPIO2  Pin = 2
+	GPIO3  Pin = 3
+	GPIO4  Pin = 4
+	GPIO5  Pin = 5
+	GPIO6  Pin = 6
+	GPIO7  Pin = 7
+	GPIO8  Pin = 8
+	GPIO9  Pin = 9
+	GPIO10 Pin = 10
+	GPIO11 Pin = 11
+	GPIO12 Pin = 12
+	GPIO13 Pin = 13
+	GPIO14 Pin = 14
+	GPIO15 Pin = 15
+	GPIO16 Pin = 16
+	GPIO17 Pin = 17
+	GPIO18 Pin = 18
+	GPIO19 Pin = 19
+	GPIO20 Pin = 20
+	GPIO21 Pin = 21
+	GPIO26 Pin = 26
+	GPIO27 Pin = 27
+	GPIO28 Pin = 28
+	GPIO29 Pin = 29
+	GPIO30 Pin = 30
+	GPIO31 Pin = 31
+	GPIO32 Pin = 32
+	GPIO33 Pin = 33
+	GPIO34 Pin = 34
+	GPIO35 Pin = 35
+	GPIO36 Pin = 36
+	GPIO37 Pin = 37
+	GPIO38 Pin = 38
+	GPIO39 Pin = 39
+	GPIO40 Pin = 40
+	GPIO41 Pin = 41
+	GPIO42 Pin = 42
+	GPIO43 Pin = 43
+	GPIO44 Pin = 44
+	GPIO45 Pin = 45
+	GPIO46 Pin = 46
+	GPIO47 Pin = 47
+	GPIO48 Pin = 48
+)
+
+// Configure this pin with the given configuration.
+func (p Pin) Configure(config PinConfig) {
+	// Output function 256 is a special value reserved for use as a regular GPIO
+	// pin. Peripherals (SPI etc) can set a custom output function by calling
+	// lowercase configure() instead with a signal name.
+	p.configure(config, 256)
+}
+
+// configure is the same as Configure, but allows for setting a specific input
+// or output signal.
+// Signals are always routed through the GPIO matrix for simplicity. Output
+// signals are configured in FUNCx_OUT_SEL_CFG which selects a particular signal
+// to output on a given pin. Input signals are configured in FUNCy_IN_SEL_CFG,
+// which sets the pin to use for a particular input signal.
+func (p Pin) configure(config PinConfig, signal uint32) {
+	if p == NoPin {
+		// This simplifies pin configuration in peripherals such as SPI.
+		return
+	}
+
+	ioConfig := uint32(0)
+
+	// MCU_SEL: Function 1 is always GPIO
+	ioConfig |= (1 << esp.IO_MUX_GPIO_MCU_SEL_Pos)
+
+	// FUN_IE: Make this pin an input pin (always set for GPIO operation)
+	ioConfig |= esp.IO_MUX_GPIO_FUN_IE
+
+	// DRV: Set drive strength to 20 mA as a default. Pins 17 and 18 are special
+	var drive uint32
+	if p == GPIO17 || p == GPIO18 {
+		drive = 1 // 20 mA
+	} else {
+		drive = 2 // 20 mA
+	}
+	ioConfig |= (drive << esp.IO_MUX_GPIO_FUN_DRV_Pos)
+
+	// WPU/WPD: Select pull mode.
+	if config.Mode == PinInputPullup {
+		ioConfig |= esp.IO_MUX_GPIO_FUN_WPU
+	} else if config.Mode == PinInputPulldown {
+		ioConfig |= esp.IO_MUX_GPIO_FUN_WPD
+	}
+
+	// Set configuration
+	ioRegister := p.ioMuxReg()
+	ioRegister.Set(ioConfig)
+
+	switch config.Mode {
+	case PinOutput:
+		// Set the 'output enable' bit.
+		if p < 32 {
+			esp.GPIO.ENABLE_W1TS.Set(1 << p)
+		} else {
+			esp.GPIO.ENABLE1_W1TS.Set(1 << (p - 32))
+		}
+		// Set the signal to read the output value from. It can be a peripheral
+		// output signal, or the special value 256 which indicates regular GPIO
+		// usage.
+		p.outFunc().Set(signal)
+	case PinInput, PinInputPullup, PinInputPulldown:
+		// Clear the 'output enable' bit.
+		if p < 32 {
+			esp.GPIO.ENABLE_W1TC.Set(1 << p)
+		} else {
+			esp.GPIO.ENABLE1_W1TC.Set(1 << (p - 32))
+		}
+		if signal != 256 {
+			// Signal is a peripheral function (not a simple GPIO). Connect this
+			// signal to the pin.
+			// Note that outFunc and inFunc work in the opposite direction.
+			// outFunc configures a pin to use a given output signal, while
+			// inFunc specifies a pin to use to read the signal from.
+			inFunc(signal).Set(esp.GPIO_FUNC_IN_SEL_CFG_SEL | uint32(p)<<esp.GPIO_FUNC_IN_SEL_CFG_IN_SEL_Pos)
+		}
+	}
+}
+
+// ioMuxReg returns the IO_MUX_n_REG register used for configuring the io mux for
+// this pin
+func (p Pin) ioMuxReg() *volatile.Register32 {
+	return (*volatile.Register32)(unsafe.Add(unsafe.Pointer(&esp.IO_MUX.GPIO0), uintptr(p)*4))
+}
+
+// outFunc returns the FUNCx_OUT_SEL_CFG register used for configuring the
+// output function selection.
+func (p Pin) outFunc() *volatile.Register32 {
+	return (*volatile.Register32)(unsafe.Add(unsafe.Pointer(&esp.GPIO.FUNC0_OUT_SEL_CFG), uintptr(p)*4))
+}
+
+// inFunc returns the FUNCy_IN_SEL_CFG register used for configuring the input
+// function selection.
+func inFunc(signal uint32) *volatile.Register32 {
+	return (*volatile.Register32)(unsafe.Add(unsafe.Pointer(&esp.GPIO.FUNC0_IN_SEL_CFG), uintptr(signal)*4))
+}
+
+// Set the pin to high or low.
+// Warning: only use this on an output pin!
+func (p Pin) Set(value bool) {
+	if value {
+		reg, mask := p.portMaskSet()
+		reg.Set(mask)
+	} else {
+		reg, mask := p.portMaskClear()
+		reg.Set(mask)
+	}
+}
+
+// Return the register and mask to enable a given GPIO pin. This can be used to
+// implement bit-banged drivers.
+//
+// Warning: only use this on an output pin!
+func (p Pin) PortMaskSet() (*uint32, uint32) {
+	reg, mask := p.portMaskSet()
+	return &reg.Reg, mask
+}
+
+// Return the register and mask to disable a given GPIO pin. This can be used to
+// implement bit-banged drivers.
+//
+// Warning: only use this on an output pin!
+func (p Pin) PortMaskClear() (*uint32, uint32) {
+	reg, mask := p.portMaskClear()
+	return &reg.Reg, mask
+}
+
+func (p Pin) portMaskSet() (*volatile.Register32, uint32) {
+	if p < 32 {
+		return &esp.GPIO.OUT_W1TS, 1 << p
+	} else {
+		return &esp.GPIO.OUT1_W1TS, 1 << (p - 32)
+	}
+}
+
+func (p Pin) portMaskClear() (*volatile.Register32, uint32) {
+	if p < 32 {
+		return &esp.GPIO.OUT_W1TC, 1 << p
+	} else {
+		return &esp.GPIO.OUT1_W1TC, 1 << (p - 32)
+	}
+}
+
+// Get returns the current value of a GPIO pin when the pin is configured as an
+// input or as an output.
+func (p Pin) Get() bool {
+	if p < 32 {
+		return esp.GPIO.IN.Get()&(1<<p) != 0
+	} else {
+		return esp.GPIO.IN1.Get()&(1<<(p-32)) != 0
+	}
+}
+
+var DefaultUART = UART0
+
+var (
+	UART0  = &_UART0
+	_UART0 = UART{Bus: esp.UART0, Buffer: NewRingBuffer()}
+	UART1  = &_UART1
+	_UART1 = UART{Bus: esp.UART1, Buffer: NewRingBuffer()}
+	UART2  = &_UART2
+	_UART2 = UART{Bus: esp.UART2, Buffer: NewRingBuffer()}
+)
+
+type UART struct {
+	Bus    *esp.UART_Type
+	Buffer *RingBuffer
+}
+
+func (uart *UART) Configure(config UARTConfig) {
+	if config.BaudRate == 0 {
+		config.BaudRate = 115200
+	}
+	// Crystal clock source is selected by default
+	uart.Bus.CLKDIV.Set(xtalClock / config.BaudRate)
+}
+
+func (uart *UART) writeByte(b byte) error {
+	for (uart.Bus.STATUS.Get()>>16)&0xff >= 128 {
+		// Read UART_TXFIFO_CNT from the status register, which indicates how
+		// many bytes there are in the transmit buffer. Wait until there are
+		// less than 128 bytes in this buffer (the default buffer size).
+	}
+	uart.Bus.FIFO.Set(uint32(b))
+	return nil
+}
+
+func (uart *UART) flush() {}

src/machine/machine_esp32s3_spi.go

@@ -0,0 +1,460 @@
+//go:build esp32s3
+
+package machine
+
+// ESP32-S3 SPI support based on ESP-IDF HAL
+// Simple but correct implementation following spi_ll.h
+// SPI0 = hardware SPI2 (FSPI), SPI1 = hardware SPI3 (HSPI)
+// https://docs.espressif.com/projects/esp-idf/en/latest/esp32s3/api-reference/peripherals/spi_master.html
+
+import (
+	"device/esp"
+	"errors"
+	"runtime/volatile"
+	"unsafe"
+)
+
+const (
+	SPI_MODE0 = uint8(0)
+	SPI_MODE1 = uint8(1)
+	SPI_MODE2 = uint8(2)
+	SPI_MODE3 = uint8(3)
+
+	// ESP32-S3 PLL clock frequency (same as ESP32-C3)
+	pplClockFreq = 80e6
+
+	// Default SPI frequency - maximum safe speed
+	SPI_DEFAULT_FREQUENCY = 80e6 // 80MHz
+)
+
+const (
+	// IO MUX function number for SPI direct connection
+	SPI_IOMUX_FUNC = 4
+)
+
+// ESP32-S3 GPIO Matrix signal indices for SPI - CORRECTED from ESP-IDF gpio_sig_map.h
+const (
+	// SPI2 (FSPI) signals - Hardware SPI2 - CORRECT VALUES from ESP-IDF
+	SPI2_CLK_OUT_IDX = uint32(101) // FSPICLK_OUT_IDX
+	SPI2_CLK_IN_IDX  = uint32(101) // FSPICLK_IN_IDX
+	SPI2_Q_OUT_IDX   = uint32(102) // FSPIQ_OUT_IDX (MISO)
+	SPI2_Q_IN_IDX    = uint32(102) // FSPIQ_IN_IDX
+	SPI2_D_OUT_IDX   = uint32(103) // FSPID_OUT_IDX (MOSI)
+	SPI2_D_IN_IDX    = uint32(103) // FSPID_IN_IDX
+	SPI2_CS0_OUT_IDX = uint32(110) // FSPICS0_OUT_IDX
+
+	// SPI3 (HSPI) signals - Hardware SPI3 - CORRECTED from ESP-IDF gpio_sig_map.h
+	// Source: /esp-idf/components/soc/esp32s3/include/soc/gpio_sig_map.h
+	SPI3_CLK_OUT_IDX = uint32(66) // Line 136: SPI3_CLK_OUT_IDX
+	SPI3_CLK_IN_IDX  = uint32(66) // Line 135: SPI3_CLK_IN_IDX
+	SPI3_Q_OUT_IDX   = uint32(67) // Line 138: SPI3_Q_OUT_IDX (MISO)
+	SPI3_Q_IN_IDX    = uint32(67) // Line 137: SPI3_Q_IN_IDX
+	SPI3_D_OUT_IDX   = uint32(68) // Line 140: SPI3_D_OUT_IDX (MOSI)
+	SPI3_D_IN_IDX    = uint32(68) // Line 139: SPI3_D_IN_IDX
+	SPI3_CS0_OUT_IDX = uint32(71) // Line 146: SPI3_CS0_OUT_IDX
+)
+
+type SPI struct {
+	Bus   interface{}
+	busID uint8
+}
+
+var (
+	SPI0 = &SPI{Bus: esp.SPI2, busID: 2} // Primary SPI (FSPI)
+	SPI1 = &SPI{Bus: esp.SPI3, busID: 3} // Secondary SPI (HSPI)
+)
+
+type SPIConfig struct {
+	Frequency uint32
+	SCK       Pin   // Serial Clock
+	SDO       Pin   // Serial Data Out (MOSI)
+	SDI       Pin   // Serial Data In  (MISO)
+	CS        Pin   // Chip Select (optional)
+	LSBFirst  bool  // MSB is default
+	Mode      uint8 // SPI_MODE0 is default
+}
+
+// Configure and make the SPI peripheral ready to use.
+// Implementation following ESP-IDF HAL with GPIO Matrix routing
+func (spi *SPI) Configure(config SPIConfig) error {
+	// Set default
+	if config.Frequency == 0 {
+		config.Frequency = SPI_DEFAULT_FREQUENCY
+	}
+
+	switch spi.busID {
+	case 2: // SPI2 (FSPI)
+		if config.SCK == 0 {
+			config.SCK = SPI1_SCK_PIN
+		}
+		if config.SDO == 0 {
+			config.SDO = SPI1_MOSI_PIN
+		}
+		if config.SDI == 0 {
+			config.SDI = SPI1_MISO_PIN
+		}
+	case 3: // SPI3 (HSPI)
+		if config.SCK == 0 {
+			config.SCK = SPI2_SCK_PIN
+		}
+		if config.SDO == 0 {
+			config.SDO = SPI2_MOSI_PIN
+		}
+		if config.SDI == 0 {
+			config.SDI = SPI2_MISO_PIN
+		}
+	default:
+	}
+
+	// Get GPIO Matrix signal indices for this SPI bus
+	var sckOutIdx, mosiOutIdx, misoInIdx, csOutIdx uint32
+	switch spi.busID {
+	case 2: // SPI2 (FSPI)
+		sckOutIdx = SPI2_CLK_OUT_IDX
+		mosiOutIdx = SPI2_D_OUT_IDX
+		misoInIdx = SPI2_Q_IN_IDX
+		csOutIdx = SPI2_CS0_OUT_IDX
+	case 3: // SPI3 (HSPI)
+		sckOutIdx = SPI3_CLK_OUT_IDX
+		mosiOutIdx = SPI3_D_OUT_IDX
+		misoInIdx = SPI3_Q_IN_IDX
+		csOutIdx = SPI3_CS0_OUT_IDX
+	default:
+		return ErrInvalidSPIBus
+	}
+
+	// Check if we can use IO MUX direct connection for better performance
+	if isDefaultSPIPins(spi.busID, config) {
+		// Use IO MUX direct connection - better signal quality and performance
+		// Configure pins using IO MUX direct connection (SPI function)
+		if config.SCK != NoPin {
+			config.SCK.configure(PinConfig{Mode: PinOutput}, SPI_IOMUX_FUNC)
+		}
+		if config.SDO != NoPin {
+			config.SDO.configure(PinConfig{Mode: PinOutput}, SPI_IOMUX_FUNC)
+		}
+		if config.SDI != NoPin {
+			config.SDI.configure(PinConfig{Mode: PinInput}, SPI_IOMUX_FUNC)
+		}
+		if config.CS != NoPin {
+			config.CS.configure(PinConfig{Mode: PinOutput}, SPI_IOMUX_FUNC)
+		}
+	} else {
+		// Use GPIO Matrix routing - more flexible but slightly slower
+		// Configure SDI (MISO) pin
+		if config.SDI != NoPin {
+			config.SDI.Configure(PinConfig{Mode: PinInput})
+			inFunc(misoInIdx).Set(esp.GPIO_FUNC_IN_SEL_CFG_SEL | uint32(config.SDI))
+		}
+
+		// Configure SDO (MOSI) pin
+		if config.SDO != NoPin {
+			config.SDO.Configure(PinConfig{Mode: PinOutput})
+			config.SDO.outFunc().Set(mosiOutIdx)
+		}
+
+		// Configure SCK (Clock) pin
+		if config.SCK != NoPin {
+			config.SCK.Configure(PinConfig{Mode: PinOutput})
+			config.SCK.outFunc().Set(sckOutIdx)
+		}
+
+		// Configure CS (Chip Select) pin
+		if config.CS != NoPin {
+			config.CS.Configure(PinConfig{Mode: PinOutput})
+			config.CS.outFunc().Set(csOutIdx)
+		}
+	}
+
+	// Enable peripheral clock and reset
+	// Without bootloader, we need to be more explicit about clock initialization
+	switch spi.busID {
+	case 2: // Hardware SPI2 (FSPI)
+		esp.SYSTEM.SetPERIP_CLK_EN0_SPI2_CLK_EN(1)
+		esp.SYSTEM.SetPERIP_RST_EN0_SPI2_RST(1)
+		esp.SYSTEM.SetPERIP_RST_EN0_SPI2_RST(0)
+	case 3: // Hardware SPI3 (HSPI)
+		esp.SYSTEM.SetPERIP_CLK_EN0_SPI3_CLK_EN(1)
+		esp.SYSTEM.SetPERIP_RST_EN0_SPI3_RST(1)
+		esp.SYSTEM.SetPERIP_RST_EN0_SPI3_RST(0)
+	}
+
+	// Get bus handle - both SPI2 and SPI3 use SPI2_Type
+	bus, ok := spi.Bus.(*esp.SPI2_Type)
+	if !ok {
+		return ErrInvalidSPIBus
+	}
+
+	// Reset timing: cs_setup_time = 0, cs_hold_time = 0
+	bus.USER1.Set(0)
+
+	// Use all 64 bytes of the buffer
+	bus.SetUSER_USR_MISO_HIGHPART(0)
+	bus.SetUSER_USR_MOSI_HIGHPART(0)
+
+	// Disable unneeded interrupts and clear all USER bits first
+	bus.SLAVE.Set(0)
+	bus.USER.Set(0)
+
+	// Clear other important registers like ESP32-C3
+	bus.MISC.Set(0)
+	bus.CTRL.Set(0)
+	bus.CLOCK.Set(0)
+
+	// Clear data buffers like ESP32-C3
+	bus.W0.Set(0)
+	bus.W1.Set(0)
+	bus.W2.Set(0)
+	bus.W3.Set(0)
+
+	// Configure master clock gate - CRITICAL: need CLK_EN bit!
+	bus.SetCLK_GATE_CLK_EN(1)         // Enable basic SPI clock (bit 0)
+	bus.SetCLK_GATE_MST_CLK_ACTIVE(1) // Enable master clock (bit 1)
+	bus.SetCLK_GATE_MST_CLK_SEL(1)    // Select master clock (bit 2)
+
+	// Configure DMA following ESP-IDF HAL
+	// Reset DMA configuration
+	bus.DMA_CONF.Set(0)
+	// Set DMA segment transaction clear enable bits
+	bus.SetDMA_CONF_SLV_TX_SEG_TRANS_CLR_EN(1)
+	bus.SetDMA_CONF_SLV_RX_SEG_TRANS_CLR_EN(1)
+	// dma_seg_trans_en = 0 (already 0 from DMA_CONF.Set(0))
+
+	// Configure master mode
+	bus.SetUSER_USR_MOSI(1)     // Enable MOSI
+	bus.SetUSER_USR_MISO(1)     // Enable MISO
+	bus.SetUSER_DOUTDIN(1)      // Full-duplex mode
+	bus.SetCTRL_WR_BIT_ORDER(0) // MSB first
+	bus.SetCTRL_RD_BIT_ORDER(0) // MSB first
+
+	// CRITICAL: Enable clock output (from working test)
+	bus.SetMISC_CK_DIS(0) // Enable CLK output - THIS IS KEY!
+
+	// Configure SPI mode (CPOL/CPHA) following ESP-IDF HAL
+	switch config.Mode {
+	case SPI_MODE0:
+		// CPOL=0, CPHA=0 (default)
+	case SPI_MODE1:
+		bus.SetUSER_CK_OUT_EDGE(1) // CPHA=1
+	case SPI_MODE2:
+		bus.SetMISC_CK_IDLE_EDGE(1) // CPOL=1
+		bus.SetUSER_CK_OUT_EDGE(1)  // CPHA=1
+	case SPI_MODE3:
+		bus.SetMISC_CK_IDLE_EDGE(1) // CPOL=1
+	}
+
+	// Configure SPI bus clock using ESP32-C3 algorithm for better accuracy
+	bus.CLOCK.Set(freqToClockDiv(config.Frequency))
+
+	return nil
+}
+
+// Transfer writes/reads a single byte using the SPI interface.
+// Implementation following ESP-IDF HAL spi_ll_user_start with proper USER register setup
+func (spi *SPI) Transfer(w byte) (byte, error) {
+	// Both SPI2 and SPI3 use SPI2_Type
+	bus, ok := spi.Bus.(*esp.SPI2_Type)
+	if !ok {
+		return 0, errors.New("invalid SPI bus type")
+	}
+
+	// Set transfer length (8 bits = 7 in register)
+	bus.SetMS_DLEN_MS_DATA_BITLEN(7)
+
+	// Clear any pending interrupt flags BEFORE starting transaction
+	bus.SetDMA_INT_CLR_TRANS_DONE_INT_CLR(1)
+
+	// Write data to buffer (use W0 register)
+	bus.W0.Set(uint32(w))
+
+	// CRITICAL: Apply configuration before transmission (like ESP-IDF spi_ll_apply_config)
+	bus.SetCMD_UPDATE(1)
+	for bus.GetCMD_UPDATE() != 0 {
+		// Wait for config to be applied
+	}
+
+	// Start transaction following ESP-IDF HAL spi_ll_user_start
+	bus.SetCMD_USR(1)
+
+	// Wait for completion using CMD_USR flag (like ESP32-C3 approach)
+	// Hardware clears CMD_USR when transaction is complete
+	timeout := 100000
+	for bus.GetCMD_USR() != 0 && timeout > 0 {
+		timeout--
+		// Wait for CMD_USR to be cleared by hardware
+	}
+
+	if timeout == 0 {
+		return 0, errors.New("SPI transfer timeout")
+	}
+
+	// Read received data from W0 register
+	result := byte(bus.W0.Get() & 0xFF)
+	return result, nil
+}
+
+// Tx handles read/write operation for SPI interface. Since SPI is a synchronous write/read
+// interface, there must always be the same number of bytes written as bytes read.
+// This is accomplished by sending zero bits if r is bigger than w or discarding
+// the incoming data if w is bigger than r.
+// Optimized implementation ported from ESP32-C3 for better performance.
+func (spi *SPI) Tx(w, r []byte) error {
+	toTransfer := len(w)
+	if len(r) > toTransfer {
+		toTransfer = len(r)
+	}
+
+	// Get bus handle - both SPI2 and SPI3 use SPI2_Type
+	bus, ok := spi.Bus.(*esp.SPI2_Type)
+	if !ok {
+		return ErrInvalidSPIBus
+	}
+
+	for toTransfer > 0 {
+		// Chunk 64 bytes at a time.
+		chunkSize := toTransfer
+		if chunkSize > 64 {
+			chunkSize = 64
+		}
+
+		// Fill tx buffer.
+		transferWords := (*[16]volatile.Register32)(unsafe.Add(unsafe.Pointer(&bus.W0), 0))
+		if len(w) >= 64 {
+			// We can fill the entire 64-byte transfer buffer with data.
+			// This loop is slightly faster than the loop below.
+			for i := 0; i < 16; i++ {
+				word := uint32(w[i*4]) | uint32(w[i*4+1])<<8 | uint32(w[i*4+2])<<16 | uint32(w[i*4+3])<<24
+				transferWords[i].Set(word)
+			}
+		} else {
+			// We can't fill the entire transfer buffer, so we need to be a bit
+			// more careful.
+			// Note that parts of the transfer buffer that aren't used still
+			// need to be set to zero, otherwise we might be transferring
+			// garbage from a previous transmission if w is smaller than r.
+			for i := 0; i < 16; i++ {
+				var word uint32
+				if i*4+3 < len(w) {
+					word |= uint32(w[i*4+3]) << 24
+				}
+				if i*4+2 < len(w) {
+					word |= uint32(w[i*4+2]) << 16
+				}
+				if i*4+1 < len(w) {
+					word |= uint32(w[i*4+1]) << 8
+				}
+				if i*4+0 < len(w) {
+					word |= uint32(w[i*4+0]) << 0
+				}
+				transferWords[i].Set(word)
+			}
+		}
+
+		// Do the transfer.
+		bus.SetMS_DLEN_MS_DATA_BITLEN(uint32(chunkSize)*8 - 1)
+
+		bus.SetCMD_UPDATE(1)
+		for bus.GetCMD_UPDATE() != 0 {
+		}
+
+		bus.SetCMD_USR(1)
+		for bus.GetCMD_USR() != 0 {
+		}
+
+		// Read rx buffer.
+		rxSize := chunkSize
+		if rxSize > len(r) {
+			rxSize = len(r)
+		}
+		for i := 0; i < rxSize; i++ {
+			r[i] = byte(transferWords[i/4].Get() >> ((i % 4) * 8))
+		}
+
+		// Cut off some part of the output buffer so the next iteration we will
+		// only send the remaining bytes.
+		if len(w) < chunkSize {
+			w = nil
+		} else {
+			w = w[chunkSize:]
+		}
+		if len(r) < chunkSize {
+			r = nil
+		} else {
+			r = r[chunkSize:]
+		}
+		toTransfer -= chunkSize
+	}
+
+	return nil
+}
+
+// Compute the SPI bus frequency from the APB clock frequency.
+// Note: APB clock is always 80MHz on ESP32-S3, independent of CPU frequency.
+// Ported from ESP32-C3 implementation for better accuracy.
+func freqToClockDiv(hz uint32) uint32 {
+	// Use APB clock frequency (80MHz), not CPU frequency!
+	// SPI peripheral is connected to APB bus which stays at 80MHz
+	const apbFreq = pplClockFreq // 80MHz
+
+	if hz >= apbFreq { // maximum frequency
+		return 1 << 31
+	}
+	if hz < (apbFreq / (16 * 64)) { // minimum frequency
+		return 15<<18 | 63<<12 | 31<<6 | 63 // pre=15, n=63
+	}
+
+	// iterate looking for an exact match
+	// or iterate all 16 prescaler options
+	// looking for the smallest error
+	var bestPre, bestN, bestErr uint32
+	bestN = 1
+	bestErr = 0xffffffff
+	q := uint32(float32(apbFreq)/float32(hz) + float32(0.5))
+	for p := uint32(0); p < 16; p++ {
+		n := q/(p+1) - 1
+		if n < 1 { // prescaler became too large, stop enum
+			break
+		}
+		if n > 63 { // prescaler too small, skip to next
+			continue
+		}
+
+		freq := apbFreq / ((p + 1) * (n + 1))
+		if freq == hz { // exact match
+			return p<<18 | n<<12 | (n/2)<<6 | n
+		}
+
+		var err uint32
+		if freq < hz {
+			err = hz - freq
+		} else {
+			err = freq - hz
+		}
+		if err < bestErr {
+			bestErr = err
+			bestPre = p
+			bestN = n
+		}
+	}
+
+	return bestPre<<18 | bestN<<12 | (bestN/2)<<6 | bestN
+}
+
+// isDefaultSPIPins checks if the given pins match the default SPI pin configuration
+// that supports IO MUX direct connection for better performance
+func isDefaultSPIPins(busID uint8, config SPIConfig) bool {
+	switch busID {
+	case 2: // SPI2 (FSPI)
+		return config.SCK == SPI1_SCK_PIN &&
+			config.SDO == SPI1_MOSI_PIN &&
+			config.SDI == SPI1_MISO_PIN &&
+			(config.CS == SPI1_CS_PIN || config.CS == NoPin)
+	case 3: // SPI3 (HSPI)
+		return config.SCK == SPI2_SCK_PIN &&
+			config.SDO == SPI2_MOSI_PIN &&
+			config.SDI == SPI2_MISO_PIN &&
+			(config.CS == SPI2_CS_PIN || config.CS == NoPin)
+	default:
+		return false
+	}
+}

src/machine/machine_esp32xx_usb.go

@@ -0,0 +1,102 @@
+//go:build esp32s3 || esp32c3
+
+package machine
+
+import (
+	"device/esp"
+	"errors"
+)
+
+// USB Serial/JTAG Controller
+// See esp32-c3_technical_reference_manual_en.pdf
+// pg. 736
+type USB_DEVICE struct {
+	Bus *esp.USB_DEVICE_Type
+}
+
+var (
+	_USBCDC = &USB_DEVICE{
+		Bus: esp.USB_DEVICE,
+	}
+
+	USBCDC Serialer = _USBCDC
+)
+
+var (
+	errUSBWrongSize            = errors.New("USB: invalid write size")
+	errUSBCouldNotWriteAllData = errors.New("USB: could not write all data")
+	errUSBBufferEmpty          = errors.New("USB: read buffer empty")
+)
+
+type Serialer interface {
+	WriteByte(c byte) error
+	Write(data []byte) (n int, err error)
+	Configure(config UARTConfig) error
+	Buffered() int
+	ReadByte() (byte, error)
+	DTR() bool
+	RTS() bool
+}
+
+func initUSB() {}
+
+func (usbdev *USB_DEVICE) Configure(config UARTConfig) error {
+	return nil
+}
+
+func (usbdev *USB_DEVICE) WriteByte(c byte) error {
+	if usbdev.Bus.GetEP1_CONF_SERIAL_IN_EP_DATA_FREE() == 0 {
+		return errUSBCouldNotWriteAllData
+	}
+
+	usbdev.Bus.SetEP1_RDWR_BYTE(uint32(c))
+	usbdev.flush()
+
+	return nil
+}
+
+func (usbdev *USB_DEVICE) Write(data []byte) (n int, err error) {
+	if len(data) == 0 || len(data) > 64 {
+		return 0, errUSBWrongSize
+	}
+
+	for i, c := range data {
+		if usbdev.Bus.GetEP1_CONF_SERIAL_IN_EP_DATA_FREE() == 0 {
+			if i > 0 {
+				usbdev.flush()
+			}
+
+			return i, errUSBCouldNotWriteAllData
+		}
+		usbdev.Bus.SetEP1_RDWR_BYTE(uint32(c))
+	}
+
+	usbdev.flush()
+	return len(data), nil
+}
+
+func (usbdev *USB_DEVICE) Buffered() int {
+	return int(usbdev.Bus.GetEP1_CONF_SERIAL_OUT_EP_DATA_AVAIL())
+}
+
+func (usbdev *USB_DEVICE) ReadByte() (byte, error) {
+	if usbdev.Bus.GetEP1_CONF_SERIAL_OUT_EP_DATA_AVAIL() != 0 {
+		return byte(usbdev.Bus.GetEP1_RDWR_BYTE()), nil
+	}
+
+	return 0, nil
+}
+
+func (usbdev *USB_DEVICE) DTR() bool {
+	return false
+}
+
+func (usbdev *USB_DEVICE) RTS() bool {
+	return false
+}
+
+func (usbdev *USB_DEVICE) flush() {
+	usbdev.Bus.SetEP1_CONF_WR_DONE(1)
+	for usbdev.Bus.GetEP1_CONF_SERIAL_IN_EP_DATA_FREE() == 0 {
+	}
+}

src/machine/machine_nrf.go

@@ -3,7 +3,9 @@
 package machine
 
 import (
+	"device/arm"
 	"device/nrf"
+	"errors"
 	"internal/binary"
 	"runtime/interrupt"
 	"unsafe"
@@ -380,6 +382,10 @@ func (f flashBlockDevice) ReadAt(p []byte, off int64) (n int, err error) {
 // If the length of p is not long enough it will be padded with 0xFF bytes.
 // This method assumes that the destination is already erased.
 func (f flashBlockDevice) WriteAt(p []byte, off int64) (n int, err error) {
+	if len(p) == 0 {
+		return 0, nil // nothing to do (and it would fail in sd_flash_write)
+	}
+
 	if FlashDataStart()+uintptr(off)+uintptr(len(p)) > FlashDataEnd() {
 		return 0, errFlashCannotWritePastEOF
 	}
@@ -387,6 +393,35 @@ func (f flashBlockDevice) WriteAt(p []byte, off int64) (n int, err error) {
 	address := FlashDataStart() + uintptr(off)
 	padded := flashPad(p, int(f.WriteBlockSize()))
 
+	// When the SoftDevice is enabled, access to the flash is restricted and
+	// must go through the SoftDevice API.
+	if isSoftDeviceEnabled() {
+		// Call sd_flash_write, which is SVC_SOC_BASE + 9 in all the
+		// SoftDevices I've checked.
+		// Documentation:
+		// https://docs.nordicsemi.com/bundle/s140_v6.0.0_api/page/group_n_r_f_s_o_c_f_u_n_c_t_i_o_n_s.html
+		numberOfWords := len(padded) / 4 // flash access goes in 32-bit words
+		result := arm.SVCall3(0x20+9, address, &padded[0], uint32(numberOfWords))
+		if result != 0 {
+			// Could not queue flash operation? Not sure when this can
+			// happen.
+			return 0, flashError
+		}
+
+		// Wait until the SoftDevice is finished.
+		flashStatus = flashStatusBusy
+		for flashStatus == flashStatusBusy {
+			handleSoftDeviceEvents()
+		}
+
+		// Check whether the operation was successful.
+		if flashStatus != flashStatusOk {
+			flashStatus = flashStatusOk
+			return 0, flashError
+		}
+		return len(p), nil
+	}
+
 	waitWhileFlashBusy()
 
 	nrf.NVMC.SetCONFIG_WEN(nrf.NVMC_CONFIG_WEN_Wen)
@@ -428,6 +463,40 @@ func (f flashBlockDevice) EraseBlockSize() int64 {
 // supports this. The start and len parameters are in block numbers, use
 // EraseBlockSize to map addresses to blocks.
 func (f flashBlockDevice) EraseBlocks(start, len int64) error {
+	// When the SoftDevice is enabled, access to the flash is restricted and
+	// must go through the SoftDevice API.
+	if isSoftDeviceEnabled() {
+		for i := range uint32(len) {
+			flashPage := uint32(FlashDataStart())/eraseBlockSizeValue + uint32(start) + i
+
+			// Call sd_flash_page_erase, which is SVC_SOC_BASE + 8 in all the
+			// SoftDevices I've checked.
+			// Documentation:
+			// https://docs.nordicsemi.com/bundle/s140_v6.0.0_api/page/group_n_r_f_s_o_c_f_u_n_c_t_i_o_n_s.html#ga9c93dd94a138ad8b5ed3693ea38ffb3e
+			result := arm.SVCall1(0x20+8, flashPage)
+			if result != 0 {
+				// Could not queue flash operation? Not sure when this can
+				// happen.
+				return flashError
+			}
+
+			// Wait until the SoftDevice is finished.
+			flashStatus = flashStatusBusy
+			for flashStatus == flashStatusBusy {
+				handleSoftDeviceEvents()
+			}
+
+			// Check whether the operation was successful.
+			if flashStatus != flashStatusOk {
+				flashStatus = flashStatusOk
+				return flashError
+			}
+		}
+		return nil
+	}
+
+	// SoftDevice is not used or enabled. Use NVIC directly.
+
 	address := FlashDataStart() + uintptr(start*f.EraseBlockSize())
 	waitWhileFlashBusy()
 
@@ -447,3 +516,52 @@ func waitWhileFlashBusy() {
 	for nrf.NVMC.GetREADY() != nrf.NVMC_READY_READY_Ready {
 	}
 }
+
+var flashError = errors.New("machine: flash operation failed")
+
+const (
+	flashStatusOk = iota
+	flashStatusError
+	flashStatusBusy
+)
+
+var flashStatus uint8 = flashStatusOk
+
+var sdEvent uint32
+
+// Process all queued SoftDevice events. May only be called when the SoftDevice is enabled.
+//
+// Normally these are handled in the same interrupt where Bluetooth events are
+// handled. But in TinyGo, that's complicated. One option would be to put it in
+// the tinygo.org/x/bluetooth package, but that would cause a circular
+// dependency between the machine and the bluetooth package. Another would be to
+// put it here, but let the bluetooth package call handleSoftDeviceEvents, but
+// that relies on updating the bluetooth package at the same time. As a
+// compromise, these events are handled directly where they are expected (here
+// in the machine package). This works in practice since there are only very few
+// of such events (at the moment, only flash-related ones which is in the
+// machine package anyway).
+func handleSoftDeviceEvents() {
+	for {
+		var result uintptr
+		if nrf.Device == "nrf52" || nrf.Device == "nrf52840" || nrf.Device == "nrf52833" {
+			// sd_evt_get: SOC_SVC_BASE_NOT_AVAILABLE + 31
+			result = arm.SVCall1(0x2C+31, &sdEvent)
+		} else {
+			return // TODO: nrf51 etc
+		}
+		if result != 0 {
+			// Some error occured. The only possible error is
+			// NRF_ERROR_NOT_FOUND, which means there are no more events.
+			return
+		}
+
+		// The following events are the same numbers in all SoftDevices I've checked.
+		switch sdEvent {
+		case 2: // NRF_EVT_FLASH_OPERATION_SUCCESS
+			flashStatus = flashStatusOk
+		case 3: // NRF_EVT_FLASH_OPERATION_ERROR
+			flashStatus = flashStatusError
+		}
+	}
+}

src/machine/machine_nrf52.go

@@ -69,3 +69,5 @@ const eraseBlockSizeValue = 4096
 func eraseBlockSize() int64 {
 	return eraseBlockSizeValue
 }
+
+const spiMaxBufferSize = 255 // from the datasheet: TXD.MAXCNT and RXD.MAXCNT

src/machine/machine_nrf52833.go

@@ -90,3 +90,5 @@ const eraseBlockSizeValue = 4096
 func eraseBlockSize() int64 {
 	return eraseBlockSizeValue
 }
+
+const spiMaxBufferSize = 0xffff // from the datasheet: TXD.MAXCNT and RXD.MAXCNT

src/machine/machine_nrf52840.go

@@ -108,3 +108,9 @@ const eraseBlockSizeValue = 4096
 func eraseBlockSize() int64 {
 	return eraseBlockSizeValue
 }
+
+const spiMaxBufferSize = 0xffff // from the datasheet: TXD.MAXCNT and RXD.MAXCNT
+
+// ADC instance for the VDDH input pin. This pin is typically connected to USB
+// input voltage (~5V) or directly to a battery.
+var ADC_VDDH = ADC{adcVDDHPin}

src/machine/machine_nrf52840_usb.go

@@ -256,7 +256,7 @@ func initEndpoint(ep, config uint32) {
 
 // SendUSBInPacket sends a packet for USBHID (interrupt in / bulk in).
 func SendUSBInPacket(ep uint32, data []byte) bool {
-	sendUSBPacket(ep, data, 0)
+	sendUSBPacket(ep, data)
 
 	// clear transfer complete flag
 	nrf.USBD.INTENCLR.Set(nrf.USBD_INTENCLR_ENDEPOUT0 << 4)
@@ -267,33 +267,32 @@ func SendUSBInPacket(ep uint32, data []byte) bool {
 // Prevent file size increases: https://github.com/tinygo-org/tinygo/pull/998
 //
 //go:noinline
-func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
+func sendUSBPacket(ep uint32, data []byte) {
+	// Select the corresponding buffer.
 	count := len(data)
-	if 0 < int(maxsize) && int(maxsize) < count {
-		count = int(maxsize)
-	}
-
+	var buffer []byte
 	if ep == 0 {
-		copy(udd_ep_control_cache_buffer[:], data[:count])
+		buffer = udd_ep_control_cache_buffer[:]
 		if count > usb.EndpointPacketSize {
+			// The packet must be sent in chunks.
 			sendOnEP0DATADONE.offset = usb.EndpointPacketSize
-			sendOnEP0DATADONE.ptr = &udd_ep_control_cache_buffer[sendOnEP0DATADONE.offset]
+			sendOnEP0DATADONE.ptr = &udd_ep_control_cache_buffer[usb.EndpointPacketSize]
 			sendOnEP0DATADONE.count = count - usb.EndpointPacketSize
 			count = usb.EndpointPacketSize
 		}
-		sendViaEPIn(
-			ep,
-			&udd_ep_control_cache_buffer[0],
-			count,
-		)
 	} else {
-		copy(udd_ep_in_cache_buffer[ep][:], data[:count])
-		sendViaEPIn(
-			ep,
-			&udd_ep_in_cache_buffer[ep][0],
-			count,
-		)
+		buffer = udd_ep_in_cache_buffer[ep][:]
 	}
+
+	// Copy the packet to the buffer.
+	copy(buffer[:len(data)], data)
+
+	// Send the first chunk of the packet.
+	sendViaEPIn(
+		ep,
+		&buffer[0],
+		count,
+	)
 }
 
 func handleEndpointRx(ep uint32) []byte {

src/machine/machine_nrf528xx.go

@@ -49,7 +49,7 @@ func (i2c *I2C) Tx(addr uint16, w, r []byte) (err error) {
 
 	// Configure for a single shot to perform both write and read (as applicable)
 	if len(w) != 0 {
-		i2c.Bus.TXD.PTR.Set(uint32(uintptr(unsafe.Pointer(&w[0]))))
+		i2c.Bus.TXD.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(unsafe.SliceData(w)))))
 		i2c.Bus.TXD.MAXCNT.Set(uint32(len(w)))
 
 		// If no read, immediately signal stop after TX
@@ -58,7 +58,7 @@ func (i2c *I2C) Tx(addr uint16, w, r []byte) (err error) {
 		}
 	}
 	if len(r) != 0 {
-		i2c.Bus.RXD.PTR.Set(uint32(uintptr(unsafe.Pointer(&r[0]))))
+		i2c.Bus.RXD.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(unsafe.SliceData(r)))))
 		i2c.Bus.RXD.MAXCNT.Set(uint32(len(r)))
 
 		// Auto-start Rx after Tx and Stop after Rx
@@ -89,6 +89,11 @@ func (i2c *I2C) Tx(addr uint16, w, r []byte) (err error) {
 		}
 	}
 
+	// Make sure the w and r buffers stay alive until this point, so they won't
+	// be garbage collected while the buffers are used by the hardware.
+	keepAliveNoEscape(unsafe.Pointer(unsafe.SliceData(w)))
+	keepAliveNoEscape(unsafe.Pointer(unsafe.SliceData(r)))
+
 	return
 }
 
@@ -117,7 +122,7 @@ func (i2c *I2C) Listen(addr uint8) error {
 //
 // For request events, the caller MUST call `Reply` to avoid hanging the i2c bus indefinitely.
 func (i2c *I2C) WaitForEvent(buf []byte) (evt I2CTargetEvent, count int, err error) {
-	i2c.BusT.RXD.PTR.Set(uint32(uintptr(unsafe.Pointer(&buf[0]))))
+	i2c.BusT.RXD.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(unsafe.SliceData(buf)))))
 	i2c.BusT.RXD.MAXCNT.Set(uint32(len(buf)))
 
 	i2c.BusT.TASKS_PREPARERX.Set(nrf.TWIS_TASKS_PREPARERX_TASKS_PREPARERX_Trigger)
@@ -134,6 +139,10 @@ func (i2c *I2C) WaitForEvent(buf []byte) (evt I2CTargetEvent, count int, err err
 		}
 	}
 
+	// Make sure buf stays alive until this point, so it won't be garbage
+	// collected while it is used by the hardware.
+	keepAliveNoEscape(unsafe.Pointer(unsafe.SliceData(buf)))
+
 	count = 0
 	evt = I2CFinish
 	err = nil
@@ -163,7 +172,7 @@ func (i2c *I2C) WaitForEvent(buf []byte) (evt I2CTargetEvent, count int, err err
 
 // Reply supplies the response data the controller.
 func (i2c *I2C) Reply(buf []byte) error {
-	i2c.BusT.TXD.PTR.Set(uint32(uintptr(unsafe.Pointer(&buf[0]))))
+	i2c.BusT.TXD.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(unsafe.SliceData(buf)))))
 	i2c.BusT.TXD.MAXCNT.Set(uint32(len(buf)))
 
 	i2c.BusT.EVENTS_STOPPED.Set(0)
@@ -180,6 +189,10 @@ func (i2c *I2C) Reply(buf []byte) error {
 		}
 	}
 
+	// Make sure the buffer stays alive until this point, so it won't be garbage
+	// collected while it is used by the hardware.
+	keepAliveNoEscape(unsafe.Pointer(unsafe.SliceData(buf)))
+
 	i2c.BusT.EVENTS_STOPPED.Set(0)
 
 	return nil

src/machine/machine_nrf52xxx.go

@@ -12,6 +12,8 @@ func CPUFrequency() uint32 {
 	return 64000000
 }
 
+var adcVDDHPin = Pin(254) // special pin number for VDDH on the nrf52840
+
 // InitADC initializes the registers needed for ADC.
 func InitADC() {
 	// Enable ADC.
@@ -26,7 +28,7 @@ func InitADC() {
 // Samples can be 1(default), 2, 4, 8, 16, 32, 64, 128, 256 samples
 func (a *ADC) Configure(config ADCConfig) {
 	var configVal uint32 = nrf.SAADC_CH_CONFIG_RESP_Bypass<<nrf.SAADC_CH_CONFIG_RESP_Pos |
-		nrf.SAADC_CH_CONFIG_RESP_Bypass<<nrf.SAADC_CH_CONFIG_RESN_Pos |
+		nrf.SAADC_CH_CONFIG_RESN_Bypass<<nrf.SAADC_CH_CONFIG_RESN_Pos |
 		nrf.SAADC_CH_CONFIG_REFSEL_Internal<<nrf.SAADC_CH_CONFIG_REFSEL_Pos |
 		nrf.SAADC_CH_CONFIG_MODE_SE<<nrf.SAADC_CH_CONFIG_MODE_Pos
 
@@ -116,36 +118,43 @@ func (a *ADC) Configure(config ADCConfig) {
 
 // Get returns the current value of an ADC pin in the range 0..0xffff.
 func (a *ADC) Get() uint16 {
-	var pwmPin uint32
+	var adcPin uint32
 	var rawValue volatile.Register16
 
 	switch a.Pin {
 	case 2:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput0
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput0
 	case 3:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput1
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput1
 	case 4:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput2
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput2
 	case 5:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput3
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput3
 	case 28:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput4
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput4
 	case 29:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput5
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput5
 	case 30:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput6
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput6
 	case 31:
-		pwmPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput7
+		adcPin = nrf.SAADC_CH_PSELP_PSELP_AnalogInput7
+	case adcVDDHPin:
+		if Device == "nrf52840" {
+			adcPin = 0x0D // VDDHDIV5 on the nrf52840
+		} else {
+			return 0
+		}
 	default:
 		return 0
 	}
 
 	// Set pin to read.
-	nrf.SAADC.CH[0].PSELN.Set(pwmPin)
-	nrf.SAADC.CH[0].PSELP.Set(pwmPin)
+	nrf.SAADC.CH[0].PSELP.Set(adcPin)
 
 	// Destination for sample result.
-	nrf.SAADC.RESULT.PTR.Set(uint32(uintptr(unsafe.Pointer(&rawValue))))
+	// Note: rawValue doesn't need to be kept alive for the GC, since the
+	// volatile read later will force it to stay alive.
+	nrf.SAADC.RESULT.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(&rawValue))))
 	nrf.SAADC.RESULT.MAXCNT.Set(1) // One sample
 
 	// Start tasks.
@@ -301,20 +310,18 @@ func (spi *SPI) Transfer(w byte) (byte, error) {
 // padded until they fit: if len(w) > len(r) the extra bytes received will be
 // dropped and if len(w) < len(r) extra 0 bytes will be sent.
 func (spi *SPI) Tx(w, r []byte) error {
-	// Unfortunately the hardware (on the nrf52832) only supports up to 255
-	// bytes in the buffers, so if either w or r is longer than that the
-	// transfer needs to be broken up in pieces.
-	// The nrf52840 supports far larger buffers however, which isn't yet
-	// supported.
+	// Unfortunately the hardware (on the nrf52832) only supports a limited
+	// amount of bytes in the buffers (depending on the chip), so if either w or
+	// r is longer than that the transfer needs to be broken up in pieces.
 	for len(r) != 0 || len(w) != 0 {
 		// Prepare the SPI transfer: set the DMA pointers and lengths.
 		// read buffer
 		nr := uint32(len(r))
 		if nr > 0 {
-			if nr > 255 {
-				nr = 255
+			if nr > spiMaxBufferSize {
+				nr = spiMaxBufferSize
 			}
-			spi.Bus.RXD.PTR.Set(uint32(uintptr(unsafe.Pointer(&r[0]))))
+			spi.Bus.RXD.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(unsafe.SliceData(r)))))
 			r = r[nr:]
 		}
 		spi.Bus.RXD.MAXCNT.Set(nr)
@@ -322,10 +329,10 @@ func (spi *SPI) Tx(w, r []byte) error {
 		// write buffer
 		nw := uint32(len(w))
 		if nw > 0 {
-			if nw > 255 {
-				nw = 255
+			if nw > spiMaxBufferSize {
+				nw = spiMaxBufferSize
 			}
-			spi.Bus.TXD.PTR.Set(uint32(uintptr(unsafe.Pointer(&w[0]))))
+			spi.Bus.TXD.PTR.Set(uint32(unsafeNoEscape(unsafe.Pointer(unsafe.SliceData(w)))))
 			w = w[nw:]
 		}
 		spi.Bus.TXD.MAXCNT.Set(nw)
@@ -335,10 +342,16 @@ func (spi *SPI) Tx(w, r []byte) error {
 		// finished if the transfer is send-only (a common case).
 		spi.Bus.TASKS_START.Set(1)
 		for spi.Bus.EVENTS_END.Get() == 0 {
+			gosched()
 		}
 		spi.Bus.EVENTS_END.Set(0)
 	}
 
+	// Make sure the w and r buffers stay alive for the GC until this point,
+	// since they are used by the hardware but not otherwise visible.
+	keepAliveNoEscape(unsafe.Pointer(unsafe.SliceData(r)))
+	keepAliveNoEscape(unsafe.Pointer(unsafe.SliceData(w)))
+
 	return nil
 }
 

src/machine/machine_nrf_bare.go

@@ -7,3 +7,7 @@ package machine
 func GetRNG() (ret uint32, err error) {
 	return getRNG()
 }
+
+func isSoftDeviceEnabled() bool {
+	return false
+}

src/machine/machine_nrf_sd.go

@@ -11,9 +11,8 @@ import (
 
 // avoid a heap allocation in GetRNG.
 var (
-	softdeviceEnabled uint8
-	bytesAvailable    uint8
-	buf               [4]uint8
+	bytesAvailable uint8
+	buf            [4]uint8
 
 	errNoSoftDeviceSupport = errors.New("rng: softdevice not supported on this device")
 	errNotEnoughRandomData = errors.New("rng: not enough random data available")
@@ -24,9 +23,7 @@ var (
 func GetRNG() (ret uint32, err error) {
 	// First check whether the SoftDevice is enabled.
 	// sd_rand_application_bytes_available_get cannot be called when the SoftDevice is not enabled.
-	arm.SVCall1(0x12, &softdeviceEnabled) // sd_softdevice_is_enabled
-
-	if softdeviceEnabled == 0 {
+	if !isSoftDeviceEnabled() {
 		return getRNG()
 	}
 
@@ -57,3 +54,8 @@ func GetRNG() (ret uint32, err error) {
 
 	return uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24, nil
 }
+
+// This function is defined in the runtime, but we need it too.
+//
+//go:linkname isSoftDeviceEnabled runtime.isSoftDeviceEnabled
+func isSoftDeviceEnabled() bool

src/machine/machine_rp2040_usb.go

@@ -128,7 +128,7 @@ func handleUSBSetAddress(setup usb.Setup) bool {
 	const ackTimeout = 570
 
 	rp.USBCTRL_REGS.SIE_STATUS.Set(rp.USBCTRL_REGS_SIE_STATUS_ACK_REC)
-	sendUSBPacket(0, []byte{}, 0)
+	sendUSBPacket(0, []byte{})
 
 	// Wait for transfer to complete with a timeout.
 	t := timer.timeElapsed()

src/machine/machine_rp2350_usb.go

@@ -131,7 +131,7 @@ func handleUSBSetAddress(setup usb.Setup) bool {
 	const ackTimeout = 570
 
 	rp.USB.SIE_STATUS.Set(rp.USB_SIE_STATUS_ACK_REC)
-	sendUSBPacket(0, []byte{}, 0)
+	sendUSBPacket(0, []byte{})
 
 	// Wait for transfer to complete with a timeout.
 	t := timer.timeElapsed()

src/machine/machine_rp2_2040.go

@@ -124,6 +124,31 @@ const (
 	fnXIP pinFunc = 0
 )
 
+// validPins confirms that the SPI pin selection is a legitimate one
+// for the the 2040 chip.
+func (spi *SPI) validPins(config SPIConfig) error {
+	var okSDI, okSDO, okSCK bool
+	switch spi.Bus {
+	case rp.SPI0:
+		okSDI = config.SDI == 0 || config.SDI == 4 || config.SDI == 16 || config.SDI == 20
+		okSDO = config.SDO == 3 || config.SDO == 7 || config.SDO == 19 || config.SDO == 23
+		okSCK = config.SCK == 2 || config.SCK == 6 || config.SCK == 18 || config.SCK == 22
+	case rp.SPI1:
+		okSDI = config.SDI == 8 || config.SDI == 12 || config.SDI == 24 || config.SDI == 28
+		okSDO = config.SDO == 11 || config.SDO == 15 || config.SDO == 27
+		okSCK = config.SCK == 10 || config.SCK == 14 || config.SCK == 26
+	}
+	switch {
+	case !okSDI:
+		return errSPIInvalidSDI
+	case !okSDO:
+		return errSPIInvalidSDO
+	case !okSCK:
+		return errSPIInvalidSCK
+	}
+	return nil
+}
+
 // Configure configures the gpio pin as per mode.
 func (p Pin) Configure(config PinConfig) {
 	if p == NoPin {

src/machine/machine_rp2_2350a.go

@@ -2,6 +2,8 @@
 
 package machine
 
+import "device/rp"
+
 // Analog pins on RP2350a.
 const (
 	ADC0 Pin = GPIO26
@@ -12,3 +14,28 @@ const (
 	// fifth ADC channel.
 	thermADC = 30
 )
+
+// validPins confirms that the SPI pin selection is a legitimate one
+// for the the 2350a chip.
+func (spi *SPI) validPins(config SPIConfig) error {
+	var okSDI, okSDO, okSCK bool
+	switch spi.Bus {
+	case rp.SPI0:
+		okSDI = config.SDI == 0 || config.SDI == 4 || config.SDI == 16 || config.SDI == 20
+		okSDO = config.SDO == 3 || config.SDO == 7 || config.SDO == 19 || config.SDO == 23
+		okSCK = config.SCK == 2 || config.SCK == 6 || config.SCK == 18 || config.SCK == 22
+	case rp.SPI1:
+		okSDI = config.SDI == 8 || config.SDI == 12 || config.SDI == 24 || config.SDI == 28
+		okSDO = config.SDO == 11 || config.SDO == 15 || config.SDO == 27
+		okSCK = config.SCK == 10 || config.SCK == 14 || config.SCK == 26
+	}
+	switch {
+	case !okSDI:
+		return errSPIInvalidSDI
+	case !okSDO:
+		return errSPIInvalidSDO
+	case !okSCK:
+		return errSPIInvalidSCK
+	}
+	return nil
+}