fix targets list

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

@@ -40,7 +40,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Restore LLVM source cache
         uses: actions/cache/restore@v4
@@ -135,7 +135,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Build TinyGo (LLVM ${{ matrix.version }})
         run: go install -tags=llvm${{ matrix.version }}

.github/workflows/docker.yml

@@ -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

@@ -137,7 +137,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Install wasmtime
         uses: bytecodealliance/actions/wasmtime/setup@v1
@@ -181,7 +181,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Install Node.js
         uses: actions/setup-node@v4
@@ -298,7 +298,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Restore LLVM source cache
         uses: actions/cache/restore@v4

.github/workflows/nix.yml

@@ -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/windows.yml

@@ -41,7 +41,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Restore cached LLVM source
         uses: actions/cache/restore@v4
@@ -147,7 +147,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Download TinyGo build
         uses: actions/download-artifact@v4
@@ -177,7 +177,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          go-version: '1.25.7'
           cache: true
       - name: Download TinyGo build
         uses: actions/download-artifact@v4
@@ -213,7 +213,7 @@ jobs:
       - name: Install Go
         uses: actions/setup-go@v6
         with:
-          go-version: '1.25.5'
+          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 +1,15 @@
+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**

GNUmakefile

@@ -814,6 +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
@@ -896,6 +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 ($(XTENSA), 0)
@@ -917,6 +923,10 @@ ifneq ($(XTENSA), 0)
 	@$(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/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 {

builder/sizes_test.go

@@ -44,7 +44,7 @@ func TestBinarySize(t *testing.T) {
 		// microcontrollers
 		{"hifive1b", "examples/echo", 3668, 280, 0, 2244},
 		{"microbit", "examples/serial", 2694, 342, 8, 2248},
-		{"wioterminal", "examples/pininterrupt", 7187, 1489, 116, 6888},
+		{"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

compiler/compiler.go

@@ -1599,7 +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")
@@ -1681,13 +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]
@@ -1715,20 +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/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/intrinsics.go

@@ -50,19 +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

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/symbol.go

@@ -184,12 +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/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/go1.21.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:
@@ -60,22 +66,29 @@ 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 readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
 entry:
@@ -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:
@@ -107,14 +123,19 @@ 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 readonly %a.data, i32 %a.len, ptr readonly %b.data, i32 %b.len, ptr %context) unnamed_addr #2 {
 entry:
@@ -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 }

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"
+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.

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_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/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/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 = FDCAN{
+		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_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 = FDCAN{
+		Bus:             stm32.FDCAN1,
+		TxAltFuncSelect: AF9_FDCAN1_FDCAN2,
+		RxAltFuncSelect: AF9_FDCAN1_FDCAN2,
+		instance:        0,
+	}
+
+	// FDCAN2 on PD12 (TX) / PD13 (RX)
+	CAN2  = &_CAN2
+	_CAN2 = FDCAN{
+		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_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

@@ -47,9 +47,15 @@ const (
 
 // SPI pins
 const (
-	SPI_SCK_PIN = GPIO7
-	SPI_SDI_PIN = GPIO9
-	SPI_SDO_PIN = GPIO8
+	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

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
 

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

@@ -509,102 +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
-}
-
-func initUSB() {
-	// nothing to do here
-}
-
-// 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

@@ -308,5 +308,3 @@ func (uart *UART) writeByte(b byte) error {
 }
 
 func (uart *UART) flush() {}
-
-// TODO: SPI

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

@@ -400,7 +400,8 @@ func (f flashBlockDevice) WriteAt(p []byte, off int64) (n int, err error) {
 		// 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
-		result := arm.SVCall3(0x20+9, address, &p[0], uint32(len(p)))
+		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.

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_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_adc.go

@@ -19,10 +19,8 @@ var adcAref uint32
 
 // InitADC resets the ADC peripheral.
 func InitADC() {
-	rp.RESETS.RESET.SetBits(rp.RESETS_RESET_ADC)
-	rp.RESETS.RESET.ClearBits(rp.RESETS_RESET_ADC)
-	for !rp.RESETS.RESET_DONE.HasBits(rp.RESETS_RESET_ADC) {
-	}
+	resetBlock(rp.RESETS_RESET_ADC)
+	unresetBlockWait(rp.RESETS_RESET_ADC)
 	// enable ADC
 	rp.ADC.CS.Set(rp.ADC_CS_EN)
 	adcAref = 3300

src/machine/machine_rp2_i2c.go

@@ -259,10 +259,7 @@ func (i2c *I2C) init(config I2CConfig) error {
 //go:inline
 func (i2c *I2C) reset() {
 	resetVal := i2c.deinit()
-	rp.RESETS.RESET.ClearBits(resetVal)
-	// Wait until reset is done.
-	for !rp.RESETS.RESET_DONE.HasBits(resetVal) {
-	}
+	unresetBlockWait(resetVal)
 }
 
 // deinit sets reset bit for I2C. Must call reset to reenable I2C after deinit.
@@ -276,15 +273,13 @@ func (i2c *I2C) deinit() (resetVal uint32) {
 		resetVal = rp.RESETS_RESET_I2C1
 	}
 	// Perform I2C reset.
-	rp.RESETS.RESET.SetBits(resetVal)
+	resetBlock(resetVal)
 
 	return resetVal
 }
 
 // tx performs blocking write followed by read to I2C bus.
 func (i2c *I2C) tx(addr uint8, tx, rx []byte) (err error) {
-	const timeout_us = 4_000
-	deadline := ticks() + timeout_us
 	if addr >= 0x80 || isReservedI2CAddr(addr) {
 		return errInvalidTgtAddr
 	}
@@ -295,6 +290,14 @@ func (i2c *I2C) tx(addr uint8, tx, rx []byte) (err error) {
 		return nil
 	}
 
+	// Base 4ms for small register pokes.
+	// Add per-byte budget. 100us/byte is conservative at 400kHz and still ok at 100kHz for modest sizes.
+	timeout_us := uint64(4_000) + uint64(txlen+rxlen)*100
+	// Cap so it doesn't go insane:
+	timeout_us = min(timeout_us, 500_000)
+
+	deadline := ticks() + timeout_us
+
 	err = i2c.disable()
 	if err != nil {
 		return err

src/machine/machine_rp2_pwm.go

@@ -153,7 +153,7 @@ func (p *pwmGroup) Period() uint64 {
 	top := p.getWrap()
 	phc := p.getPhaseCorrect()
 	Int, frac := p.getClockDiv()
-	return (16*uint64(Int) + uint64(frac)) * uint64((top+1)*(phc+1)*1e9) / (16 * freq) // cycles = (TOP+1) * (CSRPHCorrect + 1) * (DIV_INT + DIV_FRAC/16)
+	return (16*uint64(Int) + uint64(frac)) * uint64((top+1)*(phc+1)) * uint64(1e9) / (16 * freq) // cycles = (TOP+1) * (CSRPHCorrect + 1) * (DIV_INT + DIV_FRAC/16)
 }
 
 // SetInverting sets whether to invert the output of this channel.

src/machine/machine_rp2_spi.go

@@ -212,10 +212,7 @@ func (spi *SPI) setFormat(mode uint8) {
 //go:inline
 func (spi *SPI) reset() {
 	resetVal := spi.deinit()
-	rp.RESETS.RESET.ClearBits(resetVal)
-	// Wait until reset is done.
-	for !rp.RESETS.RESET_DONE.HasBits(resetVal) {
-	}
+	unresetBlockWait(resetVal)
 }
 
 //go:inline
@@ -227,7 +224,7 @@ func (spi *SPI) deinit() (resetVal uint32) {
 		resetVal = rp.RESETS_RESET_SPI1
 	}
 	// Perform SPI reset.
-	rp.RESETS.RESET.SetBits(resetVal)
+	resetBlock(resetVal)
 	return resetVal
 }
 

src/machine/machine_rp2_uart.go

@@ -73,6 +73,27 @@ func (uart *UART) Configure(config UARTConfig) error {
 	return nil
 }
 
+// Close the UART and disable its interrupt/power use.
+func (uart *UART) Close() error {
+	uart.Interrupt.Disable()
+
+	// Disable UART.
+	uart.Bus.UARTCR.ClearBits(rp.UART0_UARTCR_UARTEN)
+
+	var resetVal uint32
+	switch {
+	case uart.Bus == rp.UART0:
+		resetVal = rp.RESETS_RESET_UART0
+	case uart.Bus == rp.UART1:
+		resetVal = rp.RESETS_RESET_UART1
+	}
+
+	// reset UART
+	resetBlock(resetVal)
+
+	return nil
+}
+
 // SetBaudRate sets the baudrate to be used for the UART.
 func (uart *UART) SetBaudRate(br uint32) {
 	div := 8 * CPUFrequency() / br
@@ -148,10 +169,8 @@ func initUART(uart *UART) {
 	}
 
 	// reset UART
-	rp.RESETS.RESET.SetBits(resetVal)
-	rp.RESETS.RESET.ClearBits(resetVal)
-	for !rp.RESETS.RESET_DONE.HasBits(resetVal) {
-	}
+	resetBlock(resetVal)
+	unresetBlockWait(resetVal)
 }
 
 // handleInterrupt should be called from the appropriate interrupt handler for

src/machine/machine_rp2_usb.go

@@ -72,19 +72,15 @@ func initEndpoint(ep, config uint32) {
 
 // SendUSBInPacket sends a packet for USB (interrupt in / bulk in).
 func SendUSBInPacket(ep uint32, data []byte) bool {
-	sendUSBPacket(ep, data, 0)
+	sendUSBPacket(ep, data)
 	return true
 }
 
 // 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) {
 	count := len(data)
-	if 0 < int(maxsize) && int(maxsize) < count {
-		count = int(maxsize)
-	}
-
 	if ep == 0 {
 		if count > usb.EndpointPacketSize {
 			count = usb.EndpointPacketSize
@@ -145,7 +141,7 @@ func setEPDataPID(ep uint32, dataOne bool) {
 }
 
 func SendZlp() {
-	sendUSBPacket(0, []byte{}, 0)
+	sendUSBPacket(0, []byte{})
 }
 
 func sendViaEPIn(ep uint32, data []byte, count int) {

src/machine/machine_stm32_adc_g0.go

@@ -0,0 +1,188 @@
+//go:build stm32g0
+
+package machine
+
+import (
+	"device/stm32"
+	"unsafe"
+)
+
+// ADC sampling time constants for STM32G0
+const (
+	ADC_SMPR_1_5   = 0x0 // 1.5 ADC clock cycles
+	ADC_SMPR_3_5   = 0x1 // 3.5 ADC clock cycles
+	ADC_SMPR_7_5   = 0x2 // 7.5 ADC clock cycles
+	ADC_SMPR_12_5  = 0x3 // 12.5 ADC clock cycles
+	ADC_SMPR_19_5  = 0x4 // 19.5 ADC clock cycles
+	ADC_SMPR_39_5  = 0x5 // 39.5 ADC clock cycles
+	ADC_SMPR_79_5  = 0x6 // 79.5 ADC clock cycles
+	ADC_SMPR_160_5 = 0x7 // 160.5 ADC clock cycles
+)
+
+// InitADC initializes the registers needed for ADC.
+func InitADC() {
+	// Enable ADC clock
+	enableAltFuncClock(unsafe.Pointer(stm32.ADC))
+
+	// Ensure ADC is disabled before configuration
+	if stm32.ADC.GetCR_ADEN() != 0 {
+		// Clear ADEN by setting ADDIS
+		stm32.ADC.SetCR_ADDIS(1)
+		// Wait for ADC to be disabled
+		for stm32.ADC.GetCR_ADEN() != 0 {
+		}
+	}
+
+	// Enable ADC voltage regulator
+	stm32.ADC.SetCR_ADVREGEN(1)
+
+	// Wait for ADC voltage regulator startup time (20us at max)
+	// Using simple busy loop - approximately 1280 cycles at 64MHz = 20us
+	for i := 0; i < 1280; i++ {
+		// nop
+	}
+
+	// Configure ADC:
+	// - 12-bit resolution (RES = 0b00)
+	// - Right alignment (ALIGN = 0)
+	// - Single conversion mode (CONT = 0)
+	// - Software trigger (EXTEN = 0b00)
+	stm32.ADC.CFGR1.Set(0)
+
+	// Set clock mode to synchronous with PCLK/2
+	stm32.ADC.SetCFGR2_CKMODE(0x1) // PCLK/2
+
+	// Set sample time to 12.5 cycles for all channels using SMP1
+	stm32.ADC.SetSMPR_SMP1(ADC_SMPR_12_5)
+
+	// Calibrate ADC
+	stm32.ADC.SetCR_ADCAL(1)
+	for stm32.ADC.GetCR_ADCAL() != 0 {
+	}
+
+	// Clear ADRDY by writing 1
+	stm32.ADC.SetISR_ADRDY(1)
+
+	// Enable ADC
+	stm32.ADC.SetCR_ADEN(1)
+
+	// Wait until ADC is ready
+	for stm32.ADC.GetISR_ADRDY() == 0 {
+	}
+}
+
+// Configure configures an ADC pin to be able to read analog data.
+func (a ADC) Configure(config ADCConfig) {
+	// Configure pin as analog input
+	a.Pin.Configure(PinConfig{Mode: PinInputAnalog})
+
+	// Set sampling time based on config
+	// Use SMP2 and set SMPSEL bit for this channel to select SMP2
+	ch := a.getChannel()
+	if ch <= 18 {
+		// Select sampling time based on config (using SMP2 for per-channel control)
+		// Map microseconds to sample cycles (at ~32MHz ADC clock after /2 prescaler)
+		// Each cycle = 1/32MHz = 31.25ns
+		var smpTime int
+		switch {
+		case config.SampleTime == 0:
+			smpTime = ADC_SMPR_79_5 // Default to 79.5 cycles for good accuracy
+		case config.SampleTime <= 1:
+			smpTime = ADC_SMPR_1_5
+		case config.SampleTime <= 2:
+			smpTime = ADC_SMPR_3_5
+		case config.SampleTime <= 3:
+			smpTime = ADC_SMPR_7_5
+		case config.SampleTime <= 4:
+			smpTime = ADC_SMPR_12_5
+		case config.SampleTime <= 5:
+			smpTime = ADC_SMPR_19_5
+		case config.SampleTime <= 10:
+			smpTime = ADC_SMPR_39_5
+		case config.SampleTime <= 20:
+			smpTime = ADC_SMPR_79_5
+		default:
+			smpTime = ADC_SMPR_160_5
+		}
+		stm32.ADC.SetSMPR_SMP2(uint32(smpTime))
+
+		// Set SMPSEL bit for this channel to use SMP2
+		stm32.ADC.SMPR.SetBits(1 << (8 + ch))
+	}
+}
+
+// Get returns the current value of a ADC pin in the range 0..0xffff.
+func (a ADC) Get() uint16 {
+	ch := a.getChannel()
+
+	// Wait until channel configuration is ready if needed
+	// (CCRDY indicates when CHSELR changes are applied)
+	for stm32.ADC.GetISR_CCRDY() != 0 {
+		stm32.ADC.SetISR_CCRDY(1) // Clear by writing 1
+	}
+
+	// Select the channel to convert using CHSELR
+	// CHSELR uses a bitfield where bit N = 1 enables channel N
+	stm32.ADC.CHSELR.Set(1 << ch)
+
+	// Wait for channel configuration ready
+	for stm32.ADC.GetISR_CCRDY() == 0 {
+	}
+	stm32.ADC.SetISR_CCRDY(1) // Clear flag
+
+	// Start conversion
+	stm32.ADC.SetCR_ADSTART(1)
+
+	// Wait for end of conversion
+	for stm32.ADC.GetISR_EOC() == 0 {
+	}
+
+	// Read the 12-bit result and scale to 16-bit
+	result := uint16(stm32.ADC.GetDR_DATA()) << 4
+
+	return result
+}
+
+// getChannel returns the ADC channel number for a given pin.
+// STM32G0B1 ADC channel mapping:
+// PA0-PA7: CH0-CH7
+// PB0-PB2: CH8-CH10
+// PB10-PB12: CH11-CH13 (some variants)
+// PC4-PC5: CH17-CH18 (some variants)
+func (a ADC) getChannel() uint8 {
+	switch a.Pin {
+	case PA0:
+		return 0
+	case PA1:
+		return 1
+	case PA2:
+		return 2
+	case PA3:
+		return 3
+	case PA4:
+		return 4
+	case PA5:
+		return 5
+	case PA6:
+		return 6
+	case PA7:
+		return 7
+	case PB0:
+		return 8
+	case PB1:
+		return 9
+	case PB2:
+		return 10
+	case PB10:
+		return 11
+	case PB11:
+		return 12
+	case PB12:
+		return 13
+	case PC4:
+		return 17
+	case PC5:
+		return 18
+	}
+	return 0
+}

src/machine/machine_stm32_exti_syscfg.go

@@ -1,4 +1,4 @@
-//go:build stm32 && !stm32f1 && !stm32l5 && !stm32wlx
+//go:build stm32 && !stm32f1 && !stm32l5 && !stm32wlx && !stm32g0
 
 package machine
 

src/machine/machine_stm32_flash.go

@@ -2,6 +2,9 @@
 
 package machine
 
+// Flash support for STM32 chips, except for STM32L0 which have a different type
+// of flash.
+
 import (
 	"device/stm32"
 

src/machine/machine_stm32_gpio_reva.go

@@ -1,4 +1,4 @@
-//go:build stm32 && !stm32l4 && !stm32l5 && !stm32wlx
+//go:build stm32 && !stm32l4 && !stm32l5 && !stm32wlx && !stm32g0
 
 package machine
 

src/machine/machine_stm32_i2c_revb.go

@@ -1,4 +1,4 @@
-//go:build stm32l5 || stm32f7 || stm32l4 || stm32l0 || stm32wlx
+//go:build stm32l5 || stm32f7 || stm32l4 || stm32l0 || stm32wlx || stm32g0
 
 package machine
 

src/machine/machine_stm32_rng.go

@@ -1,4 +1,4 @@
-//go:build stm32 && !(stm32f103 || stm32l0x1)
+//go:build stm32 && !(stm32f103 || stm32l0x1 || stm32g0)
 
 package machine
 

src/machine/machine_stm32_spi.go

@@ -1,4 +1,4 @@
-//go:build stm32 && !stm32f7x2 && !stm32l5x2
+//go:build stm32 && !stm32f7x2 && !stm32l5x2 && !stm32g0
 
 package machine
 

src/machine/machine_stm32_uart.go

@@ -1,8 +1,8 @@
-//go:build stm32
+//go:build stm32 && !stm32g0
 
 package machine
 
-// Peripheral abstraction layer for UARTs on the stm32 family.
+// Peripheral abstraction layer for UARTs on the stm32 family (except stm32g0).
 
 import (
 	"device/stm32"

src/machine/machine_stm32g0.go

@@ -0,0 +1,567 @@
+//go:build stm32g0
+
+package machine
+
+// Peripheral abstraction layer for the stm32g0
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+	"runtime/volatile"
+	"unsafe"
+)
+
+const (
+	// CPU frequency for STM32G0 (64MHz via PLL: HSI16 / 1 * 8 / 2)
+	cpuFreq = 64000000
+)
+
+func CPUFrequency() uint32 {
+	return cpuFreq
+}
+
+var deviceIDAddr = []uintptr{0x1FFF7590, 0x1FFF7594, 0x1FFF7598}
+
+// Internal use: configured speed of the APB1 and APB2 timers, this should be kept
+// in sync with any changes to runtime package which configures the oscillators
+// and clock frequencies
+const APB1_TIM_FREQ = 64e6 // 64MHz (PLL: HSI16 / 1 * 8 / 2)
+const APB2_TIM_FREQ = 64e6 // 64MHz (PLL: HSI16 / 1 * 8 / 2)
+
+const (
+	PA0  = portA + 0
+	PA1  = portA + 1
+	PA2  = portA + 2
+	PA3  = portA + 3
+	PA4  = portA + 4
+	PA5  = portA + 5
+	PA6  = portA + 6
+	PA7  = portA + 7
+	PA8  = portA + 8
+	PA9  = portA + 9
+	PA10 = portA + 10
+	PA11 = portA + 11
+	PA12 = portA + 12
+	PA13 = portA + 13
+	PA14 = portA + 14
+	PA15 = portA + 15
+
+	PB0  = portB + 0
+	PB1  = portB + 1
+	PB2  = portB + 2
+	PB3  = portB + 3
+	PB4  = portB + 4
+	PB5  = portB + 5
+	PB6  = portB + 6
+	PB7  = portB + 7
+	PB8  = portB + 8
+	PB9  = portB + 9
+	PB10 = portB + 10
+	PB11 = portB + 11
+	PB12 = portB + 12
+	PB13 = portB + 13
+	PB14 = portB + 14
+	PB15 = portB + 15
+
+	PC0  = portC + 0
+	PC1  = portC + 1
+	PC2  = portC + 2
+	PC3  = portC + 3
+	PC4  = portC + 4
+	PC5  = portC + 5
+	PC6  = portC + 6
+	PC7  = portC + 7
+	PC8  = portC + 8
+	PC9  = portC + 9
+	PC10 = portC + 10
+	PC11 = portC + 11
+	PC12 = portC + 12
+	PC13 = portC + 13
+	PC14 = portC + 14
+	PC15 = portC + 15
+
+	PD0  = portD + 0
+	PD1  = portD + 1
+	PD2  = portD + 2
+	PD3  = portD + 3
+	PD4  = portD + 4
+	PD5  = portD + 5
+	PD6  = portD + 6
+	PD7  = portD + 7
+	PD8  = portD + 8
+	PD9  = portD + 9
+	PD10 = portD + 10
+	PD11 = portD + 11
+	PD12 = portD + 12
+	PD13 = portD + 13
+	PD14 = portD + 14
+	PD15 = portD + 15
+
+	PE0  = portE + 0
+	PE1  = portE + 1
+	PE2  = portE + 2
+	PE3  = portE + 3
+	PE4  = portE + 4
+	PE5  = portE + 5
+	PE6  = portE + 6
+	PE7  = portE + 7
+	PE8  = portE + 8
+	PE9  = portE + 9
+	PE10 = portE + 10
+	PE11 = portE + 11
+	PE12 = portE + 12
+	PE13 = portE + 13
+	PE14 = portE + 14
+	PE15 = portE + 15
+
+	PF0  = portF + 0
+	PF1  = portF + 1
+	PF2  = portF + 2
+	PF3  = portF + 3
+	PF4  = portF + 4
+	PF5  = portF + 5
+	PF6  = portF + 6
+	PF7  = portF + 7
+	PF8  = portF + 8
+	PF9  = portF + 9
+	PF10 = portF + 10
+	PF11 = portF + 11
+	PF12 = portF + 12
+	PF13 = portF + 13
+	PF14 = portF + 14
+	PF15 = portF + 15
+)
+
+func (p Pin) getPort() *stm32.GPIO_Type {
+	switch p / 16 {
+	case 0:
+		return stm32.GPIOA
+	case 1:
+		return stm32.GPIOB
+	case 2:
+		return stm32.GPIOC
+	case 3:
+		return stm32.GPIOD
+	case 4:
+		return stm32.GPIOE
+	case 5:
+		return stm32.GPIOF
+	default:
+		panic("machine: unknown port")
+	}
+}
+
+// enableClock enables the clock for this desired GPIO port.
+func (p Pin) enableClock() {
+	switch p / 16 {
+	case 0:
+		stm32.RCC.SetIOPENR_GPIOAEN(1)
+	case 1:
+		stm32.RCC.SetIOPENR_GPIOBEN(1)
+	case 2:
+		stm32.RCC.SetIOPENR_GPIOCEN(1)
+	case 3:
+		stm32.RCC.SetIOPENR_GPIODEN(1)
+	case 4:
+		stm32.RCC.SetIOPENR_GPIOEEN(1)
+	case 5:
+		stm32.RCC.SetIOPENR_GPIOFEN(1)
+	default:
+		panic("machine: unknown port")
+	}
+}
+
+func (p Pin) registerInterrupt() interrupt.Interrupt {
+	pin := uint8(p) % 16
+
+	switch pin {
+	case 0:
+		return interrupt.New(stm32.IRQ_EXTI0_1, func(interrupt.Interrupt) { handlePinInterrupt(0) })
+	case 1:
+		return interrupt.New(stm32.IRQ_EXTI0_1, func(interrupt.Interrupt) { handlePinInterrupt(1) })
+	case 2:
+		return interrupt.New(stm32.IRQ_EXTI2_3, func(interrupt.Interrupt) { handlePinInterrupt(2) })
+	case 3:
+		return interrupt.New(stm32.IRQ_EXTI2_3, func(interrupt.Interrupt) { handlePinInterrupt(3) })
+	case 4:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(4) })
+	case 5:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(5) })
+	case 6:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(6) })
+	case 7:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(7) })
+	case 8:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(8) })
+	case 9:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(9) })
+	case 10:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(10) })
+	case 11:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(11) })
+	case 12:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(12) })
+	case 13:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(13) })
+	case 14:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(14) })
+	case 15:
+		return interrupt.New(stm32.IRQ_EXTI4_15, func(interrupt.Interrupt) { handlePinInterrupt(15) })
+	}
+
+	return interrupt.Interrupt{}
+}
+
+//---------- UART related types and code
+
+// Configure the UART.
+func (uart *UART) configurePins(config UARTConfig) {
+	// enable the alternate functions on the TX and RX pins
+	config.TX.ConfigureAltFunc(PinConfig{Mode: PinModeUARTTX}, uart.TxAltFuncSelector)
+	config.RX.ConfigureAltFunc(PinConfig{Mode: PinModeUARTRX}, uart.RxAltFuncSelector)
+}
+
+// UART baudrate calc based on the bus and clockspeed
+func (uart *UART) getBaudRateDivisor(baudRate uint32) uint32 {
+	return CPUFrequency() / baudRate
+}
+
+// Register names vary by ST processor, these are for STM G0 family
+func (uart *UART) setRegisters() {
+	uart.rxReg = &uart.Bus.RDR
+	uart.txReg = &uart.Bus.TDR
+	uart.statusReg = &uart.Bus.ISR_FIFO_ENABLED
+	uart.txEmptyFlag = stm32.USART_ISR_TXE
+}
+
+//---------- SPI related types and code
+
+// SPI on the STM32G0 using MODER / alternate function pins
+type SPI struct {
+	Bus             *stm32.SPI_Type
+	AltFuncSelector uint8
+}
+
+func (spi *SPI) config8Bits() {
+	// Set rx threshold to 8-bits, so RXNE flag is set for 1 byte
+	spi.Bus.SetCR2_FRXTH(1)
+}
+
+// Set baud rate for SPI
+func (spi *SPI) getBaudRate(config SPIConfig) uint32 {
+	var conf uint32
+
+	localFrequency := config.Frequency
+
+	// Default
+	if localFrequency == 0 {
+		localFrequency = 4e6
+	}
+
+	// set frequency dependent on PCLK prescaler
+	switch {
+	case localFrequency < 250000:
+		conf = stm32.SPI_CR1_BR_Div256
+	case localFrequency < 500000:
+		conf = stm32.SPI_CR1_BR_Div128
+	case localFrequency < 1000000:
+		conf = stm32.SPI_CR1_BR_Div64
+	case localFrequency < 2000000:
+		conf = stm32.SPI_CR1_BR_Div32
+	case localFrequency < 4000000:
+		conf = stm32.SPI_CR1_BR_Div16
+	case localFrequency < 8000000:
+		conf = stm32.SPI_CR1_BR_Div8
+	case localFrequency < 16000000:
+		conf = stm32.SPI_CR1_BR_Div4
+	case localFrequency < 32000000:
+		conf = stm32.SPI_CR1_BR_Div2
+	default:
+		// None of the specific baudrates were selected; choose the lowest speed
+		conf = stm32.SPI_CR1_BR_Div256
+	}
+
+	return conf << stm32.SPI_CR1_BR_Pos
+}
+
+// Configure SPI pins for input output and clock
+func (spi *SPI) configurePins(config SPIConfig) {
+	config.SCK.ConfigureAltFunc(PinConfig{Mode: PinModeSPICLK}, spi.AltFuncSelector)
+	config.SDO.ConfigureAltFunc(PinConfig{Mode: PinModeSPISDO}, spi.AltFuncSelector)
+	config.SDI.ConfigureAltFunc(PinConfig{Mode: PinModeSPISDI}, spi.AltFuncSelector)
+}
+
+//---------- I2C related types and code
+
+// Gets the value for TIMINGR register
+func (i2c *I2C) getFreqRange(br uint32) uint32 {
+	// These are 'magic' values calculated by STM32CubeMX
+	// for 64MHz PCLK1 (PLL: HSI16 / 1 * 8 / 2).
+	// TODO: Do calculations based on PCLK1
+	switch br {
+	case 10 * KHz:
+		return 0xF010F3FE // 64MHz, 10kHz I2C
+	case 100 * KHz:
+		return 0x30A0A7FB // 64MHz, 100kHz I2C (Standard mode)
+	case 400 * KHz:
+		return 0x10802D9B // 64MHz, 400kHz I2C (Fast mode)
+	case 500 * KHz:
+		return 0x00802172 // 64MHz, 500kHz I2C
+	default:
+		return 0
+	}
+}
+
+// Enable peripheral clock
+func enableAltFuncClock(bus unsafe.Pointer) {
+	switch bus {
+	case unsafe.Pointer(stm32.PWR): // Power interface clock enable
+		stm32.RCC.SetAPBENR1_PWREN(1)
+	case unsafe.Pointer(stm32.I2C1): // I2C1 clock enable
+		stm32.RCC.SetAPBENR1_I2C1EN(1)
+	case unsafe.Pointer(stm32.I2C2): // I2C2 clock enable
+		stm32.RCC.SetAPBENR1_I2C2EN(1)
+	case unsafe.Pointer(stm32.USART2): // USART2 clock enable
+		stm32.RCC.SetAPBENR1_USART2EN(1)
+	case unsafe.Pointer(stm32.USART3): // USART3 clock enable
+		stm32.RCC.SetAPBENR1_USART3EN(1)
+	case unsafe.Pointer(stm32.USART4): // USART4 clock enable
+		stm32.RCC.SetAPBENR1_USART4EN(1)
+	case unsafe.Pointer(stm32.SPI2): // SPI2 clock enable
+		stm32.RCC.SetAPBENR1_SPI2EN(1)
+	case unsafe.Pointer(stm32.WWDG): // Window watchdog clock enable
+		stm32.RCC.SetAPBENR1_WWDGEN(1)
+	case unsafe.Pointer(stm32.TIM2): // TIM2 clock enable
+		stm32.RCC.SetAPBENR1_TIM2EN(1)
+	case unsafe.Pointer(stm32.TIM3): // TIM3 clock enable
+		stm32.RCC.SetAPBENR1_TIM3EN(1)
+	case unsafe.Pointer(stm32.TIM6): // TIM6 clock enable
+		stm32.RCC.SetAPBENR1_TIM6EN(1)
+	case unsafe.Pointer(stm32.TIM7): // TIM7 clock enable
+		stm32.RCC.SetAPBENR1_TIM7EN(1)
+	case unsafe.Pointer(stm32.LPUART1): // LPUART1 clock enable
+		stm32.RCC.SetAPBENR1_LPUART1EN(1)
+	case unsafe.Pointer(stm32.TIM1): // TIM1 clock enable
+		stm32.RCC.SetAPBENR2_TIM1EN(1)
+	case unsafe.Pointer(stm32.SPI1): // SPI1 clock enable
+		stm32.RCC.SetAPBENR2_SPI1EN(1)
+	case unsafe.Pointer(stm32.USART1): // USART1 clock enable
+		stm32.RCC.SetAPBENR2_USART1EN(1)
+	case unsafe.Pointer(stm32.TIM14): // TIM14 clock enable
+		stm32.RCC.SetAPBENR2_TIM14EN(1)
+	case unsafe.Pointer(stm32.TIM15): // TIM15 clock enable
+		stm32.RCC.SetAPBENR2_TIM15EN(1)
+	case unsafe.Pointer(stm32.TIM16): // TIM16 clock enable
+		stm32.RCC.SetAPBENR2_TIM16EN(1)
+	case unsafe.Pointer(stm32.TIM17): // TIM17 clock enable
+		stm32.RCC.SetAPBENR2_TIM17EN(1)
+	case unsafe.Pointer(stm32.ADC): // ADC clock enable
+		stm32.RCC.SetAPBENR2_ADCEN(1)
+	case unsafe.Pointer(stm32.FDCAN1), unsafe.Pointer(stm32.FDCAN2): // FDCAN clock enable
+		stm32.RCC.SetAPBENR1_FDCANEN(1)
+	}
+}
+
+//---------- Timer related code
+
+// Alternate function constants for STM32G0
+const (
+	AF0_SYSTEM                    = 0
+	AF1_TIM1_TIM2_TIM3_LPTIM1     = 1
+	AF2_TIM1_TIM2_TIM3_TIM14_I2C2 = 2
+	AF3_USART5_USART6_LPUART2     = 3
+	AF3_FDCAN1_FDCAN2             = 3 // FDCAN on PC2/PC3/PC4/PC5, PD12/PD13/PD14/PD15
+	AF4_USART1_USART2_TIM14       = 4
+	AF5_SPI1_SPI2_TIM16_TIM17     = 5
+	AF6_SPI2_USART3_USART4_I2C1   = 6
+	AF7_USART1_USART2_COMP1_COMP2 = 7
+	AF8_I2C1_I2C2_UCPD1_UCPD2     = 8
+	AF9_SPI2_TIM14_TIM15          = 9
+	AF9_FDCAN1_FDCAN2             = 9 // FDCAN on PA11/PA12, PB8/PB9
+)
+
+var (
+	TIM1 = TIM{
+		EnableRegister: &stm32.RCC.APBENR2,
+		EnableFlag:     stm32.RCC_APBENR2_TIM1EN,
+		Device:         stm32.TIM1,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA8, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{{PA9, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{{PA10, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{{PA11, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+		},
+		busFreq: APB2_TIM_FREQ,
+	}
+
+	TIM2 = TIM{
+		EnableRegister: &stm32.RCC.APBENR1,
+		EnableFlag:     stm32.RCC_APBENR1_TIM2EN,
+		Device:         stm32.TIM2,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA0, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}, {PA5, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}, {PA15, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{{PA1, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}, {PB3, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{{PA2, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}, {PB10, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{{PA3, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}, {PB11, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+		},
+		busFreq: APB1_TIM_FREQ,
+	}
+
+	TIM3 = TIM{
+		EnableRegister: &stm32.RCC.APBENR1,
+		EnableFlag:     stm32.RCC_APBENR1_TIM3EN,
+		Device:         stm32.TIM3,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA6, AF1_TIM1_TIM2_TIM3_LPTIM1}, {PB4, AF1_TIM1_TIM2_TIM3_LPTIM1}, {PC6, AF1_TIM1_TIM2_TIM3_LPTIM1}}},
+			{Pins: []PinFunction{{PA7, AF1_TIM1_TIM2_TIM3_LPTIM1}, {PB5, AF1_TIM1_TIM2_TIM3_LPTIM1}, {PC7, AF1_TIM1_TIM2_TIM3_LPTIM1}}},
+			{Pins: []PinFunction{{PB0, AF1_TIM1_TIM2_TIM3_LPTIM1}, {PC8, AF1_TIM1_TIM2_TIM3_LPTIM1}}},
+			{Pins: []PinFunction{{PB1, AF1_TIM1_TIM2_TIM3_LPTIM1}, {PC9, AF1_TIM1_TIM2_TIM3_LPTIM1}}},
+		},
+		busFreq: APB1_TIM_FREQ,
+	}
+
+	TIM6 = TIM{
+		EnableRegister: &stm32.RCC.APBENR1,
+		EnableFlag:     stm32.RCC_APBENR1_TIM6EN,
+		Device:         stm32.TIM6,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+		},
+		busFreq: APB1_TIM_FREQ,
+	}
+
+	TIM7 = TIM{
+		EnableRegister: &stm32.RCC.APBENR1,
+		EnableFlag:     stm32.RCC_APBENR1_TIM7EN,
+		Device:         stm32.TIM7,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+		},
+		busFreq: APB1_TIM_FREQ,
+	}
+
+	TIM14 = TIM{
+		EnableRegister: &stm32.RCC.APBENR2,
+		EnableFlag:     stm32.RCC_APBENR2_TIM14EN,
+		Device:         stm32.TIM14,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA4, AF4_USART1_USART2_TIM14}, {PA7, AF4_USART1_USART2_TIM14}, {PB1, AF0_SYSTEM}}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+		},
+		busFreq: APB2_TIM_FREQ,
+	}
+
+	TIM15 = TIM{
+		EnableRegister: &stm32.RCC.APBENR2,
+		EnableFlag:     stm32.RCC_APBENR2_TIM15EN,
+		Device:         stm32.TIM15,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA2, AF5_SPI1_SPI2_TIM16_TIM17}, {PB14, AF5_SPI1_SPI2_TIM16_TIM17}}},
+			{Pins: []PinFunction{{PA3, AF5_SPI1_SPI2_TIM16_TIM17}, {PB15, AF5_SPI1_SPI2_TIM16_TIM17}}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+		},
+		busFreq: APB2_TIM_FREQ,
+	}
+
+	TIM16 = TIM{
+		EnableRegister: &stm32.RCC.APBENR2,
+		EnableFlag:     stm32.RCC_APBENR2_TIM16EN,
+		Device:         stm32.TIM16,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA6, AF5_SPI1_SPI2_TIM16_TIM17}, {PB8, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+		},
+		busFreq: APB2_TIM_FREQ,
+	}
+
+	TIM17 = TIM{
+		EnableRegister: &stm32.RCC.APBENR2,
+		EnableFlag:     stm32.RCC_APBENR2_TIM17EN,
+		Device:         stm32.TIM17,
+		Channels: [4]TimerChannel{
+			{Pins: []PinFunction{{PA7, AF5_SPI1_SPI2_TIM16_TIM17}, {PB9, AF2_TIM1_TIM2_TIM3_TIM14_I2C2}}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+			{Pins: []PinFunction{}},
+		},
+		busFreq: APB2_TIM_FREQ,
+	}
+)
+
+func (t *TIM) registerUPInterrupt() interrupt.Interrupt {
+	switch t {
+	case &TIM1:
+		return interrupt.New(stm32.IRQ_TIM1_BRK_UP_TRG_COM, TIM1.handleUPInterrupt)
+	case &TIM2:
+		return interrupt.New(stm32.IRQ_TIM2, TIM2.handleUPInterrupt)
+	case &TIM3:
+		return interrupt.New(stm32.IRQ_TIM3_TIM4, TIM3.handleUPInterrupt)
+	case &TIM6:
+		return interrupt.New(stm32.IRQ_TIM6_DAC, TIM6.handleUPInterrupt)
+	case &TIM7:
+		return interrupt.New(stm32.IRQ_TIM7, TIM7.handleUPInterrupt)
+	case &TIM14:
+		return interrupt.New(stm32.IRQ_TIM14, TIM14.handleUPInterrupt)
+	case &TIM15:
+		return interrupt.New(stm32.IRQ_TIM15, TIM15.handleUPInterrupt)
+	case &TIM16:
+		return interrupt.New(stm32.IRQ_TIM16, TIM16.handleUPInterrupt)
+	case &TIM17:
+		return interrupt.New(stm32.IRQ_TIM17, TIM17.handleUPInterrupt)
+	}
+
+	return interrupt.Interrupt{}
+}
+
+func (t *TIM) registerOCInterrupt() interrupt.Interrupt {
+	switch t {
+	case &TIM1:
+		return interrupt.New(stm32.IRQ_TIM1_CC, TIM1.handleOCInterrupt)
+	case &TIM2:
+		return interrupt.New(stm32.IRQ_TIM2, TIM2.handleOCInterrupt)
+	case &TIM3:
+		return interrupt.New(stm32.IRQ_TIM3_TIM4, TIM3.handleOCInterrupt)
+	case &TIM6:
+		return interrupt.New(stm32.IRQ_TIM6_DAC, TIM6.handleOCInterrupt)
+	case &TIM7:
+		return interrupt.New(stm32.IRQ_TIM7, TIM7.handleOCInterrupt)
+	case &TIM14:
+		return interrupt.New(stm32.IRQ_TIM14, TIM14.handleOCInterrupt)
+	case &TIM15:
+		return interrupt.New(stm32.IRQ_TIM15, TIM15.handleOCInterrupt)
+	case &TIM16:
+		return interrupt.New(stm32.IRQ_TIM16, TIM16.handleOCInterrupt)
+	case &TIM17:
+		return interrupt.New(stm32.IRQ_TIM17, TIM17.handleOCInterrupt)
+	}
+
+	return interrupt.Interrupt{}
+}
+
+func (t *TIM) enableMainOutput() {
+	t.Device.SetBDTR_MOE(1)
+}
+
+type arrtype = uint32
+type arrRegType = volatile.Register32
+
+const (
+	ARR_MAX = 0x10000
+	PSC_MAX = 0x10000
+)
+
+func initRNG() {
+	// STM32G0B1 does not have a hardware RNG peripheral
+	// RNG is available on some other STM32G0 variants
+}

src/machine/machine_stm32g0_can.go

@@ -0,0 +1,711 @@
+//go:build stm32g0b1
+
+package machine
+
+import (
+	"device/stm32"
+	"errors"
+	"runtime/interrupt"
+	"unsafe"
+)
+
+// FDCAN Message RAM configuration
+// STM32G0B1 SRAMCAN base address: 0x4000B400
+// Each FDCAN instance has its own message RAM area
+const (
+	sramcanBase = 0x4000B400
+
+	// Message RAM layout sizes (matching STM32 HAL)
+	sramcanFLSNbr = 28 // Max. Filter List Standard Number
+	sramcanFLENbr = 8  // Max. Filter List Extended Number
+	sramcanRF0Nbr = 3  // RX FIFO 0 Elements Number
+	sramcanRF1Nbr = 3  // RX FIFO 1 Elements Number
+	sramcanTEFNbr = 3  // TX Event FIFO Elements Number
+	sramcanTFQNbr = 3  // TX FIFO/Queue Elements Number
+
+	// Element sizes in bytes
+	sramcanFLSSize = 1 * 4  // Filter Standard Element Size
+	sramcanFLESize = 2 * 4  // Filter Extended Element Size
+	sramcanRF0Size = 18 * 4 // RX FIFO 0 Element Size (for 64-byte data)
+	sramcanRF1Size = 18 * 4 // RX FIFO 1 Element Size
+	sramcanTEFSize = 2 * 4  // TX Event FIFO Element Size
+	sramcanTFQSize = 18 * 4 // TX FIFO/Queue Element Size
+
+	// Start addresses (offsets from base)
+	sramcanFLSSA = 0
+	sramcanFLESA = sramcanFLSSA + (sramcanFLSNbr * sramcanFLSSize)
+	sramcanRF0SA = sramcanFLESA + (sramcanFLENbr * sramcanFLESize)
+	sramcanRF1SA = sramcanRF0SA + (sramcanRF0Nbr * sramcanRF0Size)
+	sramcanTEFSA = sramcanRF1SA + (sramcanRF1Nbr * sramcanRF1Size)
+	sramcanTFQSA = sramcanTEFSA + (sramcanTEFNbr * sramcanTEFSize)
+	sramcanSize  = sramcanTFQSA + (sramcanTFQNbr * sramcanTFQSize)
+)
+
+// FDCAN element masks (for parsing message RAM)
+const (
+	fdcanElementMaskSTDID = 0x1FFC0000 // Standard Identifier
+	fdcanElementMaskEXTID = 0x1FFFFFFF // Extended Identifier
+	fdcanElementMaskRTR   = 0x20000000 // Remote Transmission Request
+	fdcanElementMaskXTD   = 0x40000000 // Extended Identifier flag
+	fdcanElementMaskESI   = 0x80000000 // Error State Indicator
+	fdcanElementMaskTS    = 0x0000FFFF // Timestamp
+	fdcanElementMaskDLC   = 0x000F0000 // Data Length Code
+	fdcanElementMaskBRS   = 0x00100000 // Bit Rate Switch
+	fdcanElementMaskFDF   = 0x00200000 // FD Format
+	fdcanElementMaskEFC   = 0x00800000 // Event FIFO Control
+	fdcanElementMaskMM    = 0xFF000000 // Message Marker
+	fdcanElementMaskFIDX  = 0x7F000000 // Filter Index
+	fdcanElementMaskANMF  = 0x80000000 // Accepted Non-matching Frame
+)
+
+// Interrupt flags
+const (
+	FDCAN_IT_RX_FIFO0_NEW_MESSAGE = 0x00000001
+	FDCAN_IT_RX_FIFO0_FULL        = 0x00000002
+	FDCAN_IT_RX_FIFO0_MSG_LOST    = 0x00000004
+	FDCAN_IT_RX_FIFO1_NEW_MESSAGE = 0x00000010
+	FDCAN_IT_RX_FIFO1_FULL        = 0x00000020
+	FDCAN_IT_RX_FIFO1_MSG_LOST    = 0x00000040
+	FDCAN_IT_TX_COMPLETE          = 0x00000200
+	FDCAN_IT_TX_ABORT_COMPLETE    = 0x00000400
+	FDCAN_IT_TX_FIFO_EMPTY        = 0x00000800
+	FDCAN_IT_BUS_OFF              = 0x02000000
+	FDCAN_IT_ERROR_WARNING        = 0x01000000
+	FDCAN_IT_ERROR_PASSIVE        = 0x00800000
+)
+
+// FDCAN represents an FDCAN peripheral
+type FDCAN struct {
+	Bus             *stm32.FDCAN_Type
+	TxAltFuncSelect uint8
+	RxAltFuncSelect uint8
+	Interrupt       interrupt.Interrupt
+	instance        uint8
+}
+
+// FDCANTransferRate represents CAN bus transfer rates
+type FDCANTransferRate uint32
+
+const (
+	FDCANTransferRate125kbps  FDCANTransferRate = 125000
+	FDCANTransferRate250kbps  FDCANTransferRate = 250000
+	FDCANTransferRate500kbps  FDCANTransferRate = 500000
+	FDCANTransferRate1000kbps FDCANTransferRate = 1000000
+	FDCANTransferRate2000kbps FDCANTransferRate = 2000000 // FD only
+	FDCANTransferRate4000kbps FDCANTransferRate = 4000000 // FD only
+)
+
+// FDCANMode represents the FDCAN operating mode
+type FDCANMode uint8
+
+const (
+	FDCANModeNormal           FDCANMode = 0
+	FDCANModeBusMonitoring    FDCANMode = 1
+	FDCANModeInternalLoopback FDCANMode = 2
+	FDCANModeExternalLoopback FDCANMode = 3
+)
+
+// FDCANConfig holds FDCAN configuration parameters
+type FDCANConfig struct {
+	TransferRate   FDCANTransferRate // Nominal bit rate (arbitration phase)
+	TransferRateFD FDCANTransferRate // Data bit rate (data phase), must be >= TransferRate
+	Mode           FDCANMode
+	Tx             Pin
+	Rx             Pin
+	Standby        Pin // Optional standby pin for CAN transceiver (set to NoPin if not used)
+}
+
+// FDCANTxBufferElement represents a transmit buffer element
+type FDCANTxBufferElement struct {
+	ESI bool     // Error State Indicator
+	XTD bool     // Extended ID flag
+	RTR bool     // Remote Transmission Request
+	ID  uint32   // CAN identifier (11-bit or 29-bit)
+	MM  uint8    // Message Marker
+	EFC bool     // Event FIFO Control
+	FDF bool     // FD Frame indicator
+	BRS bool     // Bit Rate Switch
+	DLC uint8    // Data Length Code (0-15)
+	DB  [64]byte // Data buffer
+}
+
+// FDCANRxBufferElement represents a receive buffer element
+type FDCANRxBufferElement struct {
+	ESI  bool     // Error State Indicator
+	XTD  bool     // Extended ID flag
+	RTR  bool     // Remote Transmission Request
+	ID   uint32   // CAN identifier
+	ANMF bool     // Accepted Non-matching Frame
+	FIDX uint8    // Filter Index
+	FDF  bool     // FD Frame
+	BRS  bool     // Bit Rate Switch
+	DLC  uint8    // Data Length Code
+	RXTS uint16   // RX Timestamp
+	DB   [64]byte // Data buffer
+}
+
+// FDCANFilterConfig represents a filter configuration
+type FDCANFilterConfig struct {
+	Index        uint8  // Filter index (0-27 for standard, 0-7 for extended)
+	Type         uint8  // 0=Range, 1=Dual, 2=Classic (ID/Mask)
+	Config       uint8  // 0=Disable, 1=FIFO0, 2=FIFO1, 3=Reject
+	ID1          uint32 // First ID or filter
+	ID2          uint32 // Second ID or mask
+	IsExtendedID bool   // true for 29-bit ID, false for 11-bit
+}
+
+var (
+	errFDCANInvalidTransferRate   = errors.New("FDCAN: invalid TransferRate")
+	errFDCANInvalidTransferRateFD = errors.New("FDCAN: invalid TransferRateFD")
+	errFDCANTimeout               = errors.New("FDCAN: timeout")
+	errFDCANTxFifoFull            = errors.New("FDCAN: Tx FIFO full")
+	errFDCANRxFifoEmpty           = errors.New("FDCAN: Rx FIFO empty")
+	errFDCANNotStarted            = errors.New("FDCAN: not started")
+)
+
+// 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}
+
+// Configure initializes the FDCAN peripheral
+func (can *FDCAN) Configure(config FDCANConfig) error {
+	// Configure standby pin if specified (for CAN transceivers with standby control)
+	// Setting it low enables the transceiver
+	if config.Standby != NoPin {
+		config.Standby.Configure(PinConfig{Mode: PinOutput})
+		config.Standby.Low()
+	}
+
+	// Enable FDCAN clock
+	enableFDCANClock()
+
+	// Configure TX and RX pins
+	config.Tx.ConfigureAltFunc(PinConfig{Mode: PinOutput}, can.TxAltFuncSelect)
+	config.Rx.ConfigureAltFunc(PinConfig{Mode: PinInputFloating}, can.RxAltFuncSelect)
+
+	// Exit from sleep mode
+	can.Bus.SetCCCR_CSR(0)
+
+	// Wait for sleep mode exit
+	timeout := 10000
+	for can.Bus.GetCCCR_CSA() != 0 {
+		timeout--
+		if timeout == 0 {
+			return errFDCANTimeout
+		}
+	}
+
+	// Request initialization
+	can.Bus.SetCCCR_INIT(1)
+
+	// Wait for init mode
+	timeout = 10000
+	for can.Bus.GetCCCR_INIT() == 0 {
+		timeout--
+		if timeout == 0 {
+			return errFDCANTimeout
+		}
+	}
+
+	// Enable configuration change
+	can.Bus.SetCCCR_CCE(1)
+
+	// Configure clock divider (only for FDCAN1)
+	if can.Bus == stm32.FDCAN1 {
+		can.Bus.SetCKDIV_PDIV(0)
+		//can.Bus.CKDIV.Set(0) // No division
+	}
+
+	// Enable automatic retransmission
+	can.Bus.SetCCCR_DAR(0)
+
+	// Disable transmit pause
+	can.Bus.SetCCCR_TXP(0)
+
+	// Enable protocol exception handling
+	can.Bus.SetCCCR_PXHD(0)
+
+	// Enable FD mode with bit rate switching
+	can.Bus.SetCCCR_FDOE(1)
+	can.Bus.SetCCCR_BRSE(1)
+
+	// Configure operating mode
+	can.Bus.SetCCCR_TEST(0)
+	can.Bus.SetCCCR_MON(0)
+	can.Bus.SetCCCR_ASM(0)
+	can.Bus.SetTEST_LBCK(0)
+
+	switch config.Mode {
+	case FDCANModeBusMonitoring:
+		can.Bus.SetCCCR_MON(1)
+	case FDCANModeInternalLoopback:
+		can.Bus.SetCCCR_TEST(1)
+		can.Bus.SetCCCR_MON(1)
+		can.Bus.SetTEST_LBCK(1)
+	case FDCANModeExternalLoopback:
+		can.Bus.SetCCCR_TEST(1)
+		can.Bus.SetTEST_LBCK(1)
+	}
+
+	// Set nominal bit timing
+	// STM32G0 runs at 64MHz, FDCAN clock = PCLK = 64MHz
+	// Bit time = (1 + NTSEG1 + NTSEG2) * tq
+	// tq = (NBRP + 1) / fCAN_CLK
+	if config.TransferRate == 0 {
+		config.TransferRate = FDCANTransferRate500kbps
+	}
+
+	nbrp, ntseg1, ntseg2, nsjw, err := can.calculateNominalBitTiming(config.TransferRate)
+	if err != nil {
+		return err
+	}
+	can.Bus.NBTP.Set(((nsjw - 1) << 25) | ((nbrp - 1) << 16) | ((ntseg1 - 1) << 8) | (ntseg2 - 1))
+
+	// Set data bit timing (for FD mode)
+	if config.TransferRateFD == 0 {
+		config.TransferRateFD = FDCANTransferRate1000kbps
+	}
+	if config.TransferRateFD < config.TransferRate {
+		return errFDCANInvalidTransferRateFD
+	}
+
+	dbrp, dtseg1, dtseg2, dsjw, err := can.calculateDataBitTiming(config.TransferRateFD)
+	if err != nil {
+		return err
+	}
+	can.Bus.DBTP.Set(((dbrp - 1) << 16) | ((dtseg1 - 1) << 8) | ((dtseg2 - 1) << 4) | (dsjw - 1))
+
+	// Configure message RAM
+	can.configureMessageRAM()
+
+	return nil
+}
+
+// Start enables the FDCAN peripheral for communication
+func (can *FDCAN) Start() error {
+	// Disable configuration change
+	can.Bus.SetCCCR_CCE(0)
+
+	// Exit initialization mode
+	can.Bus.SetCCCR_INIT(0)
+
+	// Wait for normal operation
+	timeout := 10000
+
+	for can.Bus.GetCCCR_INIT() != 0 {
+		timeout--
+		if timeout == 0 {
+			return errFDCANTimeout
+		}
+	}
+
+	return nil
+}
+
+// Stop disables the FDCAN peripheral
+func (can *FDCAN) Stop() error {
+	// Request initialization
+	can.Bus.SetCCCR_INIT(1)
+
+	// Wait for init mode
+	timeout := 10000
+	for can.Bus.GetCCCR_INIT() == 0 {
+		timeout--
+		if timeout == 0 {
+			return errFDCANTimeout
+		}
+	}
+
+	// Enable configuration change
+	can.Bus.SetCCCR_CCE(1)
+
+	return nil
+}
+
+// TxFifoIsFull returns true if the TX FIFO is full
+func (can *FDCAN) TxFifoIsFull() bool {
+	return (can.Bus.TXFQS.Get() & 0x00200000) != 0 // TFQF bit
+}
+
+// TxFifoFreeLevel returns the number of free TX FIFO elements
+func (can *FDCAN) TxFifoFreeLevel() int {
+	return int(can.Bus.TXFQS.Get() & 0x07) // TFFL[2:0]
+}
+
+// RxFifoSize returns the number of messages in RX FIFO 0
+func (can *FDCAN) RxFifoSize() int {
+	return int(can.Bus.RXF0S.Get() & 0x0F) // F0FL[3:0]
+}
+
+// RxFifoIsEmpty returns true if RX FIFO 0 is empty
+func (can *FDCAN) RxFifoIsEmpty() bool {
+	return (can.Bus.RXF0S.Get() & 0x0F) == 0
+}
+
+// TxRaw transmits a CAN frame using the raw buffer element structure
+func (can *FDCAN) TxRaw(e *FDCANTxBufferElement) error {
+	// Check if TX FIFO is full
+	if can.TxFifoIsFull() {
+		return errFDCANTxFifoFull
+	}
+
+	// Get put index
+	putIndex := (can.Bus.TXFQS.Get() >> 16) & 0x03 // TFQPI[1:0]
+
+	// Calculate TX buffer address
+	sramBase := can.getSRAMBase()
+	txAddress := sramBase + sramcanTFQSA + (uintptr(putIndex) * sramcanTFQSize)
+
+	// Build first word
+	var w1 uint32
+	id := e.ID
+	if !e.XTD {
+		// Standard ID - shift to bits [28:18]
+		id = (id & 0x7FF) << 18
+	}
+	w1 = id & 0x1FFFFFFF
+	if e.ESI {
+		w1 |= fdcanElementMaskESI
+	}
+	if e.XTD {
+		w1 |= fdcanElementMaskXTD
+	}
+	if e.RTR {
+		w1 |= fdcanElementMaskRTR
+	}
+
+	// Build second word
+	var w2 uint32
+	w2 = uint32(e.DLC) << 16
+	if e.FDF {
+		w2 |= fdcanElementMaskFDF
+	}
+	if e.BRS {
+		w2 |= fdcanElementMaskBRS
+	}
+	if e.EFC {
+		w2 |= fdcanElementMaskEFC
+	}
+	w2 |= uint32(e.MM) << 24
+
+	// Write to message RAM
+	*(*uint32)(unsafe.Pointer(txAddress)) = w1
+	*(*uint32)(unsafe.Pointer(txAddress + 4)) = w2
+
+	// Copy data bytes - must use 32-bit word access on Cortex-M0+
+	dataLen := dlcToBytes[e.DLC&0x0F]
+	numWords := (dataLen + 3) / 4
+	for w := byte(0); w < numWords; w++ {
+		var word uint32
+		baseIdx := w * 4
+		for b := byte(0); b < 4 && baseIdx+b < dataLen; b++ {
+			word |= uint32(e.DB[baseIdx+b]) << (b * 8)
+		}
+		*(*uint32)(unsafe.Pointer(txAddress + 8 + uintptr(w)*4)) = word
+	}
+
+	// Request transmission
+	can.Bus.TXBAR.Set(1 << putIndex)
+
+	return nil
+}
+
+// Tx transmits a CAN frame with the specified ID and data
+func (can *FDCAN) Tx(id uint32, data []byte, isFD, isExtendedID bool) error {
+	length := byte(len(data))
+	if length > 64 {
+		length = 64
+	}
+	if !isFD && length > 8 {
+		length = 8
+	}
+
+	e := FDCANTxBufferElement{
+		ESI: false,
+		XTD: isExtendedID,
+		RTR: false,
+		ID:  id,
+		MM:  0,
+		EFC: false,
+		FDF: isFD,
+		BRS: isFD,
+		DLC: FDCANLengthToDlc(length, isFD),
+	}
+
+	for i := byte(0); i < length; i++ {
+		e.DB[i] = data[i]
+	}
+
+	return can.TxRaw(&e)
+}
+
+// RxRaw receives a CAN frame into the raw buffer element structure
+func (can *FDCAN) RxRaw(e *FDCANRxBufferElement) error {
+	if can.RxFifoIsEmpty() {
+		return errFDCANRxFifoEmpty
+	}
+
+	// Get get index
+	getIndex := (can.Bus.RXF0S.Get() >> 8) & 0x03 // F0GI[1:0]
+
+	// Calculate RX buffer address
+	sramBase := can.getSRAMBase()
+	rxAddress := sramBase + sramcanRF0SA + (uintptr(getIndex) * sramcanRF0Size)
+
+	// Read first word
+	w1 := *(*uint32)(unsafe.Pointer(rxAddress))
+	e.ESI = (w1 & fdcanElementMaskESI) != 0
+	e.XTD = (w1 & fdcanElementMaskXTD) != 0
+	e.RTR = (w1 & fdcanElementMaskRTR) != 0
+
+	if e.XTD {
+		e.ID = w1 & fdcanElementMaskEXTID
+	} else {
+		e.ID = (w1 & fdcanElementMaskSTDID) >> 18
+	}
+
+	// Read second word
+	w2 := *(*uint32)(unsafe.Pointer(rxAddress + 4))
+	e.RXTS = uint16(w2 & fdcanElementMaskTS)
+	e.DLC = uint8((w2 & fdcanElementMaskDLC) >> 16)
+	e.BRS = (w2 & fdcanElementMaskBRS) != 0
+	e.FDF = (w2 & fdcanElementMaskFDF) != 0
+	e.FIDX = uint8((w2 & fdcanElementMaskFIDX) >> 24)
+	e.ANMF = (w2 & fdcanElementMaskANMF) != 0
+
+	// Copy data bytes - must use 32-bit word access on Cortex-M0+
+	dataLen := dlcToBytes[e.DLC&0x0F]
+	numWords := (dataLen + 3) / 4
+	for w := byte(0); w < numWords; w++ {
+		word := *(*uint32)(unsafe.Pointer(rxAddress + 8 + uintptr(w)*4))
+		baseIdx := w * 4
+		for b := byte(0); b < 4 && baseIdx+b < dataLen; b++ {
+			e.DB[baseIdx+b] = byte(word >> (b * 8))
+		}
+	}
+
+	// Acknowledge the read
+	can.Bus.RXF0A.Set(uint32(getIndex))
+
+	return nil
+}
+
+// Rx receives a CAN frame and returns its components
+func (can *FDCAN) Rx() (id uint32, dlc byte, data []byte, isFD, isExtendedID bool, err error) {
+	e := FDCANRxBufferElement{}
+	err = can.RxRaw(&e)
+	if err != nil {
+		return 0, 0, nil, false, false, err
+	}
+
+	length := FDCANDlcToLength(e.DLC, e.FDF)
+	return e.ID, length, e.DB[:length], e.FDF, e.XTD, nil
+}
+
+// SetInterrupt configures interrupt handling for the FDCAN peripheral
+func (can *FDCAN) SetInterrupt(ie uint32, callback func(*FDCAN)) error {
+	if callback == nil {
+		can.Bus.IE.ClearBits(ie)
+		return nil
+	}
+
+	can.Bus.IE.SetBits(ie)
+
+	idx := can.instance
+	fdcanInstances[idx] = can
+
+	for i := uint(0); i < 32; i++ {
+		if ie&(1<<i) != 0 {
+			fdcanCallbacks[idx][i] = callback
+		}
+	}
+
+	can.Interrupt.Enable()
+	return nil
+}
+
+// ConfigureFilter configures a message filter
+func (can *FDCAN) ConfigureFilter(config FDCANFilterConfig) error {
+	sramBase := can.getSRAMBase()
+
+	if config.IsExtendedID {
+		// Extended filter
+		if config.Index >= sramcanFLENbr {
+			return errors.New("FDCAN: filter index out of range")
+		}
+
+		filterAddr := sramBase + sramcanFLESA + (uintptr(config.Index) * sramcanFLESize)
+
+		// Build filter elements
+		w1 := (uint32(config.Config) << 29) | (config.ID1 & 0x1FFFFFFF)
+		w2 := (uint32(config.Type) << 30) | (config.ID2 & 0x1FFFFFFF)
+
+		*(*uint32)(unsafe.Pointer(filterAddr)) = w1
+		*(*uint32)(unsafe.Pointer(filterAddr + 4)) = w2
+	} else {
+		// Standard filter
+		if config.Index >= sramcanFLSNbr {
+			return errors.New("FDCAN: filter index out of range")
+		}
+
+		filterAddr := sramBase + sramcanFLSSA + (uintptr(config.Index) * sramcanFLSSize)
+
+		// Build filter element
+		w := (uint32(config.Type) << 30) |
+			(uint32(config.Config) << 27) |
+			((config.ID1 & 0x7FF) << 16) |
+			(config.ID2 & 0x7FF)
+
+		*(*uint32)(unsafe.Pointer(filterAddr)) = w
+	}
+
+	return nil
+}
+
+func (can *FDCAN) getSRAMBase() uintptr {
+	base := uintptr(sramcanBase)
+	if can.Bus == stm32.FDCAN2 {
+		base += sramcanSize
+	}
+	return base
+}
+
+func (can *FDCAN) configureMessageRAM() {
+	sramBase := can.getSRAMBase()
+
+	// Clear message RAM
+	for addr := sramBase; addr < sramBase+sramcanSize; addr += 4 {
+		*(*uint32)(unsafe.Pointer(addr)) = 0
+	}
+
+	// Configure filter counts (using RXGFC register)
+	// LSS = number of standard filters, LSE = number of extended filters
+	rxgfc := can.Bus.RXGFC.Get()
+	rxgfc &= ^uint32(0xFF000000)            // Clear LSS and LSE
+	rxgfc |= (sramcanFLSNbr << 24)          // Standard filters
+	rxgfc |= (sramcanFLENbr << 24) & 0xFF00 // Extended filters (shifted)
+	can.Bus.RXGFC.Set(rxgfc)
+}
+
+func (can *FDCAN) calculateNominalBitTiming(rate FDCANTransferRate) (brp, tseg1, tseg2, sjw uint32, err error) {
+	// STM32G0 FDCAN clock = 64MHz
+	// Target: 80% sample point
+	// Bit time = (1 + TSEG1 + TSEG2) time quanta
+	switch rate {
+	case FDCANTransferRate125kbps:
+		// 64MHz / 32 = 2MHz, 16 tq per bit = 125kbps
+		return 32, 13, 2, 4, nil
+	case FDCANTransferRate250kbps:
+		// 64MHz / 16 = 4MHz, 16 tq per bit = 250kbps
+		return 16, 13, 2, 4, nil
+	case FDCANTransferRate500kbps:
+		// 64MHz / 8 = 8MHz, 16 tq per bit = 500kbps
+		return 8, 13, 2, 4, nil
+	case FDCANTransferRate1000kbps:
+		// 64MHz / 4 = 16MHz, 16 tq per bit = 1Mbps
+		return 4, 13, 2, 4, nil
+	default:
+		return 0, 0, 0, 0, errFDCANInvalidTransferRate
+	}
+}
+
+func (can *FDCAN) calculateDataBitTiming(rate FDCANTransferRate) (brp, tseg1, tseg2, sjw uint32, err error) {
+	// STM32G0 FDCAN clock = 64MHz
+	// For data phase, we need higher bit rates
+	switch rate {
+	case FDCANTransferRate125kbps:
+		return 32, 13, 2, 4, nil
+	case FDCANTransferRate250kbps:
+		return 16, 13, 2, 4, nil
+	case FDCANTransferRate500kbps:
+		return 8, 13, 2, 4, nil
+	case FDCANTransferRate1000kbps:
+		return 4, 13, 2, 4, nil
+	case FDCANTransferRate2000kbps:
+		// 64MHz / 2 = 32MHz, 16 tq per bit = 2Mbps
+		return 2, 13, 2, 4, nil
+	case FDCANTransferRate4000kbps:
+		// 64MHz / 1 = 64MHz, 16 tq per bit = 4Mbps
+		return 1, 13, 2, 4, nil
+	default:
+		return 0, 0, 0, 0, errFDCANInvalidTransferRateFD
+	}
+}
+
+// FDCANDlcToLength converts a DLC value to actual byte length
+func FDCANDlcToLength(dlc byte, isFD bool) byte {
+	if dlc > 15 {
+		dlc = 15
+	}
+	length := dlcToBytes[dlc]
+	if !isFD && length > 8 {
+		return 8
+	}
+	return length
+}
+
+// FDCANLengthToDlc converts a byte length to DLC value
+func FDCANLengthToDlc(length byte, isFD bool) byte {
+	if !isFD {
+		if length > 8 {
+			return 8
+		}
+		return length
+	}
+
+	switch {
+	case length <= 8:
+		return length
+	case length <= 12:
+		return 9
+	case length <= 16:
+		return 10
+	case length <= 20:
+		return 11
+	case length <= 24:
+		return 12
+	case length <= 32:
+		return 13
+	case length <= 48:
+		return 14
+	default:
+		return 15
+	}
+}
+
+// Interrupt handling
+var (
+	fdcanInstances [2]*FDCAN
+	fdcanCallbacks [2][32]func(*FDCAN)
+)
+
+func fdcanHandleInterrupt(idx int) {
+	if fdcanInstances[idx] == nil {
+		return
+	}
+
+	can := fdcanInstances[idx]
+	ir := can.Bus.IR.Get()
+	can.Bus.IR.Set(ir) // Clear interrupt flags
+
+	for i := uint(0); i < 32; i++ {
+		if ir&(1<<i) != 0 && fdcanCallbacks[idx][i] != nil {
+			fdcanCallbacks[idx][i](can)
+		}
+	}
+}
+
+// Data returns the received data as a slice
+func (e *FDCANRxBufferElement) Data() []byte {
+	return e.DB[:FDCANDlcToLength(e.DLC, e.FDF)]
+}
+
+// Length returns the actual data length
+func (e *FDCANRxBufferElement) Length() byte {
+	return FDCANDlcToLength(e.DLC, e.FDF)
+}
+
+// enableFDCANClock enables the FDCAN peripheral clock
+func enableFDCANClock() {
+	// FDCAN clock is on APB1
+	stm32.RCC.SetAPBENR1_FDCANEN(1)
+}

src/machine/machine_stm32g0_gpio.go

@@ -0,0 +1,92 @@
+//go:build stm32g0
+
+package machine
+
+import (
+	"device/stm32"
+)
+
+// This variant of the GPIO input interrupt logic is for
+// STM32G0 chips which use a different EXTI register structure
+// with IMR1, RTSR1, FTSR1, and separate RPR1/FPR1 pending registers.
+
+// Callbacks for pin interrupt events
+var pinCallbacks [16]func(Pin)
+
+// The pin currently associated with interrupt callback
+// for a given slot.
+var interruptPins [16]Pin
+
+// SetInterrupt sets an interrupt to be executed when a particular pin changes
+// state. The pin should already be configured as an input, including a pull up
+// or down if no external pull is provided.
+//
+// This call will replace a previously set callback on this pin. You can pass a
+// nil func to unset the pin change interrupt. If you do so, the change
+// parameter is ignored and can be set to any value (such as 0).
+func (p Pin) SetInterrupt(change PinChange, callback func(Pin)) error {
+	port := uint32(uint8(p) / 16)
+	pin := uint8(p) % 16
+
+	enableEXTIConfigRegisters()
+
+	if callback == nil {
+		stm32.EXTI.IMR1.ClearBits(1 << pin)
+		pinCallbacks[pin] = nil
+		return nil
+	}
+
+	if pinCallbacks[pin] != nil {
+		// The pin was already configured.
+		// To properly re-configure a pin, unset it first and set a new
+		// configuration.
+		return ErrNoPinChangeChannel
+	}
+
+	// Set the callback now (before the interrupt is enabled) to avoid
+	// possible race condition
+	pinCallbacks[pin] = callback
+	interruptPins[pin] = p
+
+	crReg := getEXTIConfigRegister(pin)
+	shift := (pin & 0x3) * 4
+	crReg.ReplaceBits(port, 0xf, shift)
+
+	if (change & PinRising) != 0 {
+		stm32.EXTI.RTSR1.SetBits(1 << pin)
+	}
+	if (change & PinFalling) != 0 {
+		stm32.EXTI.FTSR1.SetBits(1 << pin)
+	}
+	stm32.EXTI.IMR1.SetBits(1 << pin)
+
+	intr := p.registerInterrupt()
+	intr.SetPriority(0)
+	intr.Enable()
+
+	return nil
+}
+
+func handlePinInterrupt(pin uint8) {
+	// STM32G0 has separate rising and falling pending registers
+	// Check both and clear the appropriate one
+	mask := uint32(1 << pin)
+	if stm32.EXTI.RPR1.HasBits(mask) {
+		// Writing 1 to the pending register clears the pending flag
+		stm32.EXTI.RPR1.Set(mask)
+
+		callback := pinCallbacks[pin]
+		if callback != nil {
+			callback(interruptPins[pin])
+		}
+	}
+	if stm32.EXTI.FPR1.HasBits(mask) {
+		// Writing 1 to the pending register clears the pending flag
+		stm32.EXTI.FPR1.Set(mask)
+
+		callback := pinCallbacks[pin]
+		if callback != nil {
+			callback(interruptPins[pin])
+		}
+	}
+}

src/machine/machine_stm32g0_spi.go

@@ -0,0 +1,99 @@
+//go:build stm32g0
+
+package machine
+
+// SPI on STM32G0 uses 16-bit registers
+
+import (
+	"device/stm32"
+	"runtime/volatile"
+	"unsafe"
+)
+
+// SPIConfig is used to store config info for SPI.
+type SPIConfig struct {
+	Frequency uint32
+	SCK       Pin
+	SDO       Pin
+	SDI       Pin
+	LSBFirst  bool
+	Mode      uint8
+}
+
+// Configure is intended to setup the STM32 SPI peripheral
+func (spi *SPI) Configure(config SPIConfig) error {
+	// disable SPI interface before any configuration changes
+	spi.Bus.CR1.ClearBits(stm32.SPI_CR1_SPE)
+
+	// enable clock for SPI
+	enableAltFuncClock(unsafe.Pointer(spi.Bus))
+
+	// init pins - use defaults if not specified
+	if config.SCK == 0 && config.SDO == 0 && config.SDI == 0 {
+		config.SCK = SPI0_SCK_PIN
+		config.SDO = SPI0_SDO_PIN
+		config.SDI = SPI0_SDI_PIN
+	}
+	spi.configurePins(config)
+
+	// Get SPI baud rate divisor
+	conf := spi.getBaudRate(config)
+
+	// set polarity and phase on the SPI interface
+	switch config.Mode {
+	case Mode1:
+		conf |= stm32.SPI_CR1_CPHA
+	case Mode2:
+		conf |= stm32.SPI_CR1_CPOL
+	case Mode3:
+		conf |= stm32.SPI_CR1_CPOL | stm32.SPI_CR1_CPHA
+	}
+
+	// set bit transfer order
+	if config.LSBFirst {
+		conf |= stm32.SPI_CR1_LSBFIRST
+	}
+
+	// set SPI master
+	conf |= stm32.SPI_CR1_MSTR | stm32.SPI_CR1_SSI
+
+	// use software CS (GPIO) by default
+	conf |= stm32.SPI_CR1_SSM
+
+	// Set CR1 configuration WITHOUT enabling SPE yet
+	// (STM32G0 requires CR2 DS bits to be set before SPE is enabled)
+	spi.Bus.CR1.Set(uint16(conf))
+
+	// Series-specific configuration to set 8-bit transfer mode (must be done before SPE)
+	spi.config8Bits()
+
+	// Now enable SPI
+	spi.Bus.SetCR1_SPE(1)
+
+	return nil
+}
+
+// Transfer writes/reads a single byte using the SPI interface.
+func (spi *SPI) Transfer(w byte) (byte, error) {
+	// STM32G0 requires 8-bit access to DR for 8-bit transfers
+	// Using 16-bit access causes data packing issues
+	dr := (*volatile.Register8)(unsafe.Pointer(&spi.Bus.DR))
+
+	// Write data to be transmitted to the SPI data register (8-bit access)
+	dr.Set(w)
+
+	// Wait until transmit complete
+	for !spi.Bus.SR.HasBits(stm32.SPI_SR_TXE) {
+	}
+
+	// Wait until receive complete
+	for !spi.Bus.SR.HasBits(stm32.SPI_SR_RXNE) {
+	}
+
+	// Wait until SPI is not busy
+	for spi.Bus.SR.HasBits(stm32.SPI_SR_BSY) {
+	}
+
+	// Return received data from SPI data register (8-bit access)
+	return dr.Get(), nil
+}

src/machine/machine_stm32g0_uart.go

@@ -0,0 +1,86 @@
+//go:build stm32g0
+
+package machine
+
+// Peripheral abstraction layer for UARTs on the stm32g0 family.
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+	"runtime/volatile"
+	"unsafe"
+)
+
+// UART representation
+type UART struct {
+	Buffer            *RingBuffer
+	Bus               *stm32.USART_Type
+	Interrupt         interrupt.Interrupt
+	TxAltFuncSelector uint8
+	RxAltFuncSelector uint8
+
+	// Registers specific to the chip
+	rxReg       *volatile.Register32
+	txReg       *volatile.Register32
+	statusReg   *volatile.Register32
+	txEmptyFlag uint32
+}
+
+// Configure the UART.
+func (uart *UART) Configure(config UARTConfig) {
+	// Default baud rate to 115200.
+	if config.BaudRate == 0 {
+		config.BaudRate = 115200
+	}
+
+	// Set the GPIO pins to defaults if they're not set
+	if config.TX == 0 && config.RX == 0 {
+		config.TX = UART_TX_PIN
+		config.RX = UART_RX_PIN
+	}
+
+	// STM32 families have different, but compatible, registers for
+	// basic UART functions.  For each family populate the registers
+	// into `uart`.
+	uart.setRegisters()
+
+	// Enable USART clock
+	enableAltFuncClock(unsafe.Pointer(uart.Bus))
+
+	uart.configurePins(config)
+
+	// Set baud rate
+	uart.SetBaudRate(config.BaudRate)
+
+	// Enable USART port, tx, rx and rx interrupts
+	// STM32G0 uses CR1_FIFO_ENABLED register
+	uart.Bus.CR1_FIFO_ENABLED.Set(stm32.USART_CR1_TE | stm32.USART_CR1_RE | stm32.USART_CR1_RXNEIE | stm32.USART_CR1_UE)
+
+	// Enable RX IRQ
+	uart.Interrupt.SetPriority(0xc0)
+	uart.Interrupt.Enable()
+}
+
+// handleInterrupt should be called from the appropriate interrupt handler for
+// this UART instance.
+func (uart *UART) handleInterrupt(interrupt.Interrupt) {
+	uart.Receive(byte((uart.rxReg.Get() & 0xFF)))
+}
+
+// SetBaudRate sets the communication speed for the UART. Defer to chip-specific
+// routines for calculation
+func (uart *UART) SetBaudRate(br uint32) {
+	divider := uart.getBaudRateDivisor(br)
+	uart.Bus.BRR.Set(divider)
+}
+
+// WriteByte writes a byte of data to the UART.
+func (uart *UART) writeByte(c byte) error {
+	uart.txReg.Set(uint32(c))
+
+	for !uart.statusReg.HasBits(uart.txEmptyFlag) {
+	}
+	return nil
+}
+
+func (uart *UART) flush() {}

src/machine/machine_stm32g0_wwdg.go

@@ -0,0 +1,173 @@
+//go:build stm32g0
+
+package machine
+
+import (
+	"device/stm32"
+	"unsafe"
+)
+
+// WindowWatchdog provides access to the Window Watchdog (WWDG) peripheral.
+// Unlike IWDG, WWDG must be refreshed within a specific window - not too early
+// and not too late. This provides protection against both runaway code and
+// code that gets stuck in a loop refreshing the watchdog.
+var WindowWatchdog = &windowWatchdogImpl{}
+
+// WindowWatchdogConfig holds configuration for the window watchdog timer.
+// The timeout (in microseconds) before the watchdog fires.
+// The valid range depends on System frequency.
+// At 64MHz: ~64µs to ~524ms
+type WindowWatchdogConfig struct {
+	TimeoutMicros uint32
+
+	// The window value as a percentage of timeout (0-100).
+	// Refresh must occur when counter is below this percentage of max.
+	// Default (0) sets window to 100% (no window restriction).
+	WindowPercent uint8
+}
+
+// WWDG prescaler values
+const (
+	wwdgPrescaler1   = 0 // CK Counter Clock (PCLK/4096) / 1
+	wwdgPrescaler2   = 1 // CK Counter Clock (PCLK/4096) / 2
+	wwdgPrescaler4   = 2 // CK Counter Clock (PCLK/4096) / 4
+	wwdgPrescaler8   = 3 // CK Counter Clock (PCLK/4096) / 8
+	wwdgPrescaler16  = 4 // CK Counter Clock (PCLK/4096) / 16
+	wwdgPrescaler32  = 5 // CK Counter Clock (PCLK/4096) / 32
+	wwdgPrescaler64  = 6 // CK Counter Clock (PCLK/4096) / 64
+	wwdgPrescaler128 = 7 // CK Counter Clock (PCLK/4096) / 128
+)
+
+// WWDG counter limits
+const (
+	wwdgCounterMin = 0x40 // Minimum counter value (T6 must be set)
+	wwdgCounterMax = 0x7F // Maximum counter value (7 bits)
+	wwdgWindowMax  = 0x7F // Maximum window value
+)
+
+type windowWatchdogImpl struct {
+	counter   uint8 // Configured counter reload value
+	prescaler uint8 // Configured prescaler
+}
+
+// Configure the window watchdog.
+//
+// This method should not be called after the watchdog is started.
+// The WWDG cannot be disabled once started, except by a system reset.
+//
+// Timeout formula: t_WWDG = (1/PCLK) × 4096 × 2^WDGTB × (T[5:0] + 1)
+// Where T[5:0] = counter value - 0x40
+// Refer RM0444 Rev 6 861/1384
+func (wd *windowWatchdogImpl) Configure(config WindowWatchdogConfig) error {
+	// Enable WWDG clock
+	enableAltFuncClock(unsafe.Pointer(stm32.WWDG))
+
+	// Calculate prescaler and counter value from timeout
+	// Base tick = PCLK / 4096
+	// With prescaler: tick = PCLK / (4096 * 2^prescaler)
+	// Timeout = tick * (counter - 0x3F)
+
+	pclk := CPUFrequency()              // Assuming PCLK = CPU frequency (no APB prescaler)
+	baseTick := (4096 * 1000000) / pclk // Base tick in nanoseconds * 1000 for precision
+
+	timeout := config.TimeoutMicros
+	if timeout == 0 {
+		timeout = 10000 // Default 10ms
+	}
+
+	// Find the best prescaler and counter-combination
+	var bestPrescaler uint8
+	var bestCounter uint8
+	found := false
+
+	for prescaler := uint8(0); prescaler <= 7; prescaler++ {
+		// Tick duration in nanoseconds * 1000
+		tickNs := baseTick << prescaler
+
+		// Counter value needed (counter - 0x3F = timeout / tick)
+		// Rearranged: counter = (timeout * 1000 / tickNs) + 0x3F
+		counterVal := (uint32(timeout) * 1000000 / tickNs) + 0x3F
+
+		if counterVal >= wwdgCounterMin && counterVal <= wwdgCounterMax {
+			bestPrescaler = prescaler
+			bestCounter = uint8(counterVal)
+			found = true
+			break
+		}
+	}
+
+	if !found {
+		// Use maximum timeout
+		bestPrescaler = wwdgPrescaler128
+		bestCounter = wwdgCounterMax
+	}
+
+	wd.prescaler = bestPrescaler
+	wd.counter = bestCounter
+
+	// Calculate window value
+	windowVal := uint8(wwdgWindowMax)
+	if config.WindowPercent > 0 && config.WindowPercent < 100 {
+		// Window = 0x40 + ((counter - 0x40) * percent / 100)
+		counterRange := uint16(bestCounter) - wwdgCounterMin
+		windowOffset := (counterRange * uint16(config.WindowPercent)) / 100
+		windowVal = uint8(wwdgCounterMin + windowOffset)
+	}
+	stm32.WWDG.CFR.Set((uint32(bestPrescaler) << stm32.WWDG_CFR_WDGTB_Pos) | uint32(windowVal))
+
+	return nil
+}
+
+// Start enables the window watchdog.
+// Once started, the WWDG cannot be disabled except by a system reset.
+func (wd *windowWatchdogImpl) Start() error {
+	stm32.WWDG.CR.Set(uint32(wd.counter) | (1 << 7))
+	return nil
+}
+
+// Update refreshes the window watchdog counter.
+// This must be called within the configured window to prevent a reset.
+// Calling too early (counter > window) or too late (counter <= 0x3F) causes reset.
+func (wd *windowWatchdogImpl) Update() {
+	stm32.WWDG.CR.Set(uint32(wd.counter) | (1 << 7))
+}
+
+// GetCounter returns the current WWDG counter value.
+// Useful for timing refresh operations within the window.
+func (wd *windowWatchdogImpl) GetCounter() uint8 {
+	return uint8(stm32.WWDG.CR.Get() & 0x7F)
+}
+
+// EnableEarlyWakeupInterrupt enables the Early Wakeup Interrupt (EWI).
+// The EWI is triggered when the counter reaches 0x40, giving the application
+// a chance to refresh the watchdog or perform cleanup before reset.
+func (wd *windowWatchdogImpl) EnableEarlyWakeupInterrupt() {
+	stm32.WWDG.CFR.SetBits(stm32.WWDG_CFR_EWI)
+}
+
+// ClearEarlyWakeupFlag clears the Early Wakeup Interrupt flag.
+// Must be called in the interrupt handler.
+func (wd *windowWatchdogImpl) ClearEarlyWakeupFlag() {
+	stm32.WWDG.SR.Set(0) // Write 0 to clear EWIF
+}
+
+// IsEarlyWakeupFlagSet returns true if the Early Wakeup Interrupt flag is set.
+func (wd *windowWatchdogImpl) IsEarlyWakeupFlagSet() bool {
+	return stm32.WWDG.SR.Get()&1 != 0
+}
+
+// GetMaxTimeout returns the maximum timeout in microseconds for the current PCLK.
+// Max timeout = (1/PCLK) × 4096 × 128 × 64
+// At 64MHz: ~524ms = 524288µs
+func (wd *windowWatchdogImpl) GetMaxTimeout() uint32 {
+	pclk := uint64(CPUFrequency())
+	return uint32((uint64(4096) * 128 * 64 * 1000000) / pclk)
+}
+
+// GetMinTimeout returns the minimum timeout in microseconds for the current PCLK.
+// Min timeout = (1/PCLK) × 4096 × 1 × 1
+// At 64MHz: ~64µs
+func (wd *windowWatchdogImpl) GetMinTimeout() uint32 {
+	pclk := uint64(CPUFrequency())
+	return uint32((uint64(4096) * 1000000) / pclk)
+}

src/machine/machine_stm32go_exti.go

@@ -0,0 +1,26 @@
+//go:build stm32g0
+
+package machine
+
+import (
+	"device/stm32"
+	"runtime/volatile"
+)
+
+func getEXTIConfigRegister(pin uint8) *volatile.Register32 {
+	switch (pin & 0xf) / 4 {
+	case 0:
+		return &stm32.EXTI.EXTICR1
+	case 1:
+		return &stm32.EXTI.EXTICR2
+	case 2:
+		return &stm32.EXTI.EXTICR3
+	case 3:
+		return &stm32.EXTI.EXTICR4
+	}
+	return nil
+}
+
+func enableEXTIConfigRegisters() {
+	// EXTI configuration is in the EXTI peripheral on STM32G0, no enable needed
+}

src/machine/machine_stm32l0_flash.go

@@ -0,0 +1,165 @@
+//go:build stm32l0
+
+package machine
+
+// The STM32L0 series of MCUs has a different type of flash than other STM32
+// series chips. The programming interface is different, and the flash is erased
+// to zero bits instead of one bits as on most flash. So this requires a
+// different implementation.
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+	"runtime/volatile"
+	"unsafe"
+)
+
+// compile-time check for ensuring we fulfill BlockDevice interface
+var _ BlockDevice = flashBlockDevice{}
+
+var Flash flashBlockDevice
+
+type flashBlockDevice struct {
+}
+
+// ReadAt reads the given number of bytes from the block device.
+func (f flashBlockDevice) ReadAt(p []byte, off int64) (n int, err error) {
+	if FlashDataStart()+uintptr(off)+uintptr(len(p)) > FlashDataEnd() {
+		return 0, errFlashCannotReadPastEOF
+	}
+
+	data := unsafe.Slice((*byte)(unsafe.Pointer(FlashDataStart()+uintptr(off))), len(p))
+	copy(p, data)
+
+	return len(p), nil
+}
+
+// WriteAt writes the given number of bytes to the block device.
+// Only word-sized (32 bits) length data can be programmed.
+// If the length of p is not long enough it will be padded with zero bytes.
+// This method assumes that the destination is already erased.
+func (f flashBlockDevice) WriteAt(p []byte, off int64) (n int, err error) {
+	if FlashDataStart()+uintptr(off)+uintptr(len(p)) > FlashDataEnd() {
+		return 0, errFlashCannotWritePastEOF
+	}
+	if uintptr(off)%4 != 0 {
+		// Offset must be aligned on a word boundary.
+		return 0, errFlashCannotWriteData
+	}
+
+	unlockFlash()
+	defer lockFlash()
+
+	// Write words in this area.
+	for i := 0; i < len(p); i += 4 {
+		// Construct the word to write.
+		word := uint32(p[i])
+		if i+1 < len(p) {
+			word |= uint32(p[i+1]) << 8
+		}
+		if i+2 < len(p) {
+			word |= uint32(p[i+2]) << 16
+		}
+		if i+3 < len(p) {
+			word |= uint32(p[i+3]) << 24
+		}
+
+		// Find the pointer address to write.
+		address := FlashDataStart() + uintptr(off) + uintptr(i)
+
+		// Write the word to flash.
+		(*volatile.Register32)(unsafe.Pointer(address)).Set(word)
+
+		// Check for any errors.
+		if stm32.FLASH.SR.Get()&(stm32.Flash_SR_WRPERR|stm32.Flash_SR_NOTZEROERR|stm32.Flash_SR_SIZERR) != 0 {
+			return i, errFlashCannotWriteData
+		}
+	}
+
+	return len(p), nil
+}
+
+// Size returns the number of bytes in this block device.
+func (f flashBlockDevice) Size() int64 {
+	return int64(FlashDataEnd() - FlashDataStart())
+}
+
+// WriteBlockSize returns the block size in which data can be written to
+// memory. It can be used by a client to optimize writes, non-aligned writes
+// should always work correctly.
+func (f flashBlockDevice) WriteBlockSize() int64 {
+	return 4
+}
+
+func eraseBlockSize() int64 {
+	return 128
+}
+
+// EraseBlockSize returns the smallest erasable area on this particular chip
+// in bytes. This is used for the block size in EraseBlocks.
+// It must be a power of two, and may be as small as 1. A typical size is 4096.
+func (f flashBlockDevice) EraseBlockSize() int64 {
+	return eraseBlockSize()
+}
+
+// EraseBlocks erases the given number of blocks. An implementation may
+// transparently coalesce ranges of blocks into larger bundles if the chip
+// supports this. The start and len parameters are in block numbers, use
+// EraseBlockSize to map addresses to blocks.
+// Note that block 0 should map to the address of FlashDataStart().
+func (f flashBlockDevice) EraseBlocks(start, len int64) error {
+	// Flash needs to be unlocked to be able to erase it.
+	unlockFlash()
+	defer lockFlash()
+
+	// Set the flash programming mode to erase a page.
+	// Note: lockFlash() will reset these flags to 0 so we don't need to
+	// explicitly set them to 0.
+	stm32.FLASH.PECR.Set(stm32.Flash_PECR_ERASE | stm32.Flash_PECR_PROG)
+
+	// Erase all pages in this range.
+	for i := uintptr(start); i < uintptr(start)+uintptr(len); i++ {
+		// Find the pointer address somewhere in the page to erase.
+		address := FlashDataStart() + i*uintptr(eraseBlockSize())
+
+		// To erase, write any value to that address.
+		(*volatile.Register32)(unsafe.Pointer(address)).Set(uint32(address))
+
+		// Check for any errors.
+		// The only error (that is not a programming error) that could happen is
+		// if a row is in a protected sector.
+		if stm32.FLASH.SR.Get()&(stm32.Flash_SR_WRPERR|stm32.Flash_SR_SIZERR) != 0 {
+			return errFlashCannotErasePage
+		}
+	}
+
+	return nil
+}
+
+func unlockFlash() {
+	// Make sure the flash peripheral clock is enabled.
+	stm32.RCC.AHBENR.SetBits(stm32.RCC_AHBENR_MIFEN)
+
+	// Wait for the flash memory not to be busy.
+	for stm32.FLASH.GetSR_BSY() != 0 {
+	}
+
+	// Disable interrupts while writing, since no memory operations may happen
+	// while the unlock sequence is ongoing.
+	mask := interrupt.Disable()
+
+	// Remove PELOCK bit.
+	stm32.FLASH.PEKEYR.Set(0x89ABCDEF)
+	stm32.FLASH.PEKEYR.Set(0x02030405)
+
+	// Remove PRGLOCK bit.
+	stm32.FLASH.PRGKEYR.Set(0x8C9DAEBF)
+	stm32.FLASH.PRGKEYR.Set(0x13141516)
+
+	interrupt.Restore(mask)
+}
+
+func lockFlash() {
+	// Set PELOCK to 1, which also automatically sets PRGLOCK to 1.
+	stm32.FLASH.PECR.Set(stm32.Flash_PECR_PELOCK)
+}

src/machine/spi.go

@@ -1,4 +1,4 @@
-//go:build !baremetal || atmega || esp32 || fe310 || k210 || nrf || (nxp && !mk66f18) || rp2040 || rp2350 || sam || (stm32 && !stm32f7x2 && !stm32l5x2)
+//go:build !baremetal || atmega || attiny85 || esp32 || fe310 || k210 || nrf || (nxp && !mk66f18) || rp2040 || rp2350 || sam || (stm32 && !stm32f7x2 && !stm32l5x2)
 
 package machine
 

src/machine/spi_tx.go

@@ -1,4 +1,4 @@
-//go:build atmega || fe310 || k210 || (nxp && !mk66f18) || (stm32 && !stm32f7x2 && !stm32l5x2)
+//go:build atmega || attiny85 || fe310 || k210 || (nxp && !mk66f18) || (stm32 && !stm32f7x2 && !stm32l5x2)
 
 // This file implements the SPI Tx function for targets that don't have a custom
 // (faster) implementation for it.

src/machine/usb.go

@@ -81,21 +81,6 @@ func usbSerial() string {
 	return ""
 }
 
-// strToUTF16LEDescriptor converts a utf8 string into a string descriptor
-// note: the following code only converts ascii characters to UTF16LE. In order
-// to do a "proper" conversion, we would need to pull in the 'unicode/utf16'
-// package, which at the time this was written added 512 bytes to the compiled
-// binary.
-func strToUTF16LEDescriptor(in string, out []byte) {
-	out[0] = byte(len(out))
-	out[1] = descriptor.TypeString
-	for i, rune := range in {
-		out[(i<<1)+2] = byte(rune)
-		out[(i<<1)+3] = 0
-	}
-	return
-}
-
 const cdcLineInfoSize = 7
 
 var (
@@ -137,51 +122,53 @@ var (
 	usbStallHandler [NumberOfUSBEndpoints]func(usb.Setup) bool
 )
 
+var usbLangInfo = [4]byte{
+	// length = 4 bytes
+	0x04,
+	// descriptor type = string
+	0x03,
+	// language codes
+	// 0x0409 = English (United States)
+	0x09, 0x04,
+}
+
 // sendDescriptor creates and sends the various USB descriptor types that
 // can be requested by the host.
 func sendDescriptor(setup usb.Setup) {
 	switch setup.WValueH {
 	case descriptor.TypeConfiguration:
-		sendUSBPacket(0, usbDescriptor.Configuration, setup.WLength)
+		sendDescriptorData(usbDescriptor.Configuration, setup.WLength)
 		return
+
 	case descriptor.TypeDevice:
 		usbDescriptor.Configure(usbVendorID(), usbProductID())
-		sendUSBPacket(0, usbDescriptor.Device, setup.WLength)
+		sendDescriptorData(usbDescriptor.Device, setup.WLength)
 		return
 
 	case descriptor.TypeString:
 		switch setup.WValueL {
 		case 0:
-			usb_trans_buffer[0] = 0x04
-			usb_trans_buffer[1] = 0x03
-			usb_trans_buffer[2] = 0x09
-			usb_trans_buffer[3] = 0x04
-			sendUSBPacket(0, usb_trans_buffer[:4], setup.WLength)
+			sendDescriptorData(usbLangInfo[:], setup.WLength)
 
 		case usb.IPRODUCT:
-			b := usb_trans_buffer[:(len(usbProduct())<<1)+2]
-			strToUTF16LEDescriptor(usbProduct(), b)
-			sendUSBPacket(0, b, setup.WLength)
+			sendDescriptorString(usbProduct(), setup.WLength)
 
 		case usb.IMANUFACTURER:
-			b := usb_trans_buffer[:(len(usbManufacturer())<<1)+2]
-			strToUTF16LEDescriptor(usbManufacturer(), b)
-			sendUSBPacket(0, b, setup.WLength)
+			sendDescriptorString(usbManufacturer(), setup.WLength)
 
 		case usb.ISERIAL:
-			sz := len(usbSerial())
-			if sz == 0 {
+			serial := usbSerial()
+			if len(serial) == 0 {
 				SendZlp()
 			} else {
-				b := usb_trans_buffer[:(sz<<1)+2]
-				strToUTF16LEDescriptor(usbSerial(), b)
-				sendUSBPacket(0, b, setup.WLength)
+				sendDescriptorString(serial, setup.WLength)
 			}
 		}
+		// TODO: why do we do this when WValueL is unknown?
 		return
 	case descriptor.TypeHIDReport:
 		if h, ok := usbDescriptor.HID[setup.WIndex]; ok {
-			sendUSBPacket(0, h, setup.WLength)
+			sendDescriptorData(h, setup.WLength)
 			return
 		}
 	case descriptor.TypeDeviceQualifier:
@@ -194,6 +181,39 @@ func sendDescriptor(setup usb.Setup) {
 	return
 }
 
+// sendDescriptorString sends a string descriptor, truncating it to fit maxLen or the buffer size.
+// note: the following code only converts ascii characters to UTF16LE. In order
+// to do a "proper" conversion, we would need to pull in the 'unicode/utf16'
+// package, which at the time this was written added 512 bytes to the compiled
+// binary.
+// TODO: old comment, re-evaluate
+func sendDescriptorString(data string, maxLen uint16) {
+	if maxLen < 2 {
+		// Something has gone horribly wrong.
+		SendZlp()
+		return
+	}
+
+	// Clamp the length.
+	maxEncBytes := min(len(usb_trans_buffer), len(udd_ep_control_cache_buffer), int(maxLen))
+	data = data[:min(len(data), (maxEncBytes-2)/2)]
+
+	// Write the header.
+	buf := usb_trans_buffer[:2*len(data)+2]
+	hdr, body := buf[:2], buf[2:]
+	hdr[0] = byte(len(buf))
+	hdr[1] = descriptor.TypeString
+
+	// Convert the string to UTF16.
+	// NOTE: Using range here would cause the length to disagree when multibyte codepoints are present.
+	for i := 0; i < len(data); i++ {
+		body[2*i] = byte(data[i])
+		body[2*i+1] = 0
+	}
+
+	sendUSBPacket(0, buf)
+}
+
 func handleStandardSetup(setup usb.Setup) bool {
 	switch setup.BRequest {
 	case usb.GET_STATUS:
@@ -206,7 +226,7 @@ func handleStandardSetup(setup usb.Setup) bool {
 			}
 		}
 
-		sendUSBPacket(0, usb_trans_buffer[:2], setup.WLength)
+		sendDescriptorData(usb_trans_buffer[:2], setup.WLength)
 		return true
 
 	case usb.CLEAR_FEATURE:
@@ -251,7 +271,7 @@ func handleStandardSetup(setup usb.Setup) bool {
 
 	case usb.GET_CONFIGURATION:
 		usb_trans_buffer[0] = usbConfiguration
-		sendUSBPacket(0, usb_trans_buffer[:1], setup.WLength)
+		sendDescriptorData(usb_trans_buffer[:1], setup.WLength)
 		return true
 
 	case usb.SET_CONFIGURATION:
@@ -271,7 +291,7 @@ func handleStandardSetup(setup usb.Setup) bool {
 
 	case usb.GET_INTERFACE:
 		usb_trans_buffer[0] = usbSetInterface
-		sendUSBPacket(0, usb_trans_buffer[:1], setup.WLength)
+		sendDescriptorData(usb_trans_buffer[:1], setup.WLength)
 		return true
 
 	case usb.SET_INTERFACE:
@@ -285,6 +305,21 @@ func handleStandardSetup(setup usb.Setup) bool {
 	}
 }
 
+// sendDescriptorData sends a descriptor, truncating it to fit maxLen or the buffer size.
+func sendDescriptorData(data []byte, maxLen uint16) {
+	data = lenToCap(data)
+	data = data[:min(len(data), len(udd_ep_control_cache_buffer), int(maxLen))]
+	sendUSBPacket(0, data)
+}
+
+// Set the cap of the slice to the length.
+// This is safe, but cannot be proven by the compiler.
+//
+//go:nobounds
+func lenToCap(b []byte) []byte {
+	return b[:len(b):len(b)]
+}
+
 func EnableCDC(txHandler func(), rxHandler func([]byte), setupHandler func(usb.Setup) bool) {
 	if len(usbDescriptor.Device) == 0 {
 		usbDescriptor = descriptor.CDC

src/os/file_anyos.go

@@ -166,6 +166,19 @@ func Chown(name string, uid, gid int) error {
 	return nil
 }
 
+// Lchown changes the numeric uid and gid of the named file.
+// If the file is a symbolic link, it changes the uid and gid of the link itself.
+// If there is an error, it will be of type [*PathError].
+//
+// If there is an error, it will be of type *PathError.
+func Lchown(name string, uid, gid int) error {
+	e := ignoringEINTR(func() error { return syscall.Lchown(name, uid, gid) })
+	if e != nil {
+		return &PathError{Op: "lchown", Path: name, Err: e}
+	}
+	return nil
+}
+
 // ignoringEINTR makes a function call and repeats it if it returns an
 // EINTR error. This appears to be required even though we install all
 // signal handlers with SA_RESTART: see #22838, #38033, #38836, #40846.

src/os/os_chmod_test.go

@@ -12,6 +12,7 @@ import (
 	"errors"
 	"io/fs"
 	. "os"
+	"path/filepath"
 	"runtime"
 	"testing"
 )
@@ -70,3 +71,38 @@ func TestChownErr(t *testing.T) {
 		}
 	}
 }
+
+func TestLchownErr(t *testing.T) {
+	if runtime.GOOS == "windows" || runtime.GOOS == "plan9" {
+		t.Log("skipping on " + runtime.GOOS)
+		return
+	}
+
+	var (
+		TEST_UID_ROOT = 0
+		TEST_GID_ROOT = 0
+	)
+
+	f := newFile("TestLchown", t)
+	defer Remove(f.Name())
+	defer f.Close()
+	link := filepath.Join(TempDir(), "TestLchownLink")
+	_ = Symlink(f.Name(), link)
+	defer Remove(link)
+
+	// EACCES
+	if err := Lchown(link, TEST_UID_ROOT, TEST_GID_ROOT); err != nil {
+		errCmp := fs.PathError{Op: "lchown", Path: link, Err: errors.New("operation not permitted")}
+		if errors.Is(err, &errCmp) {
+			t.Fatalf("lchown(%s, uid=%v, gid=%v): got '%v', want 'operation not permitted'", link, TEST_UID_ROOT, TEST_GID_ROOT, err)
+		}
+	}
+
+	// ENOENT
+	if err := Chown("invalid", Geteuid(), Getgid()); err != nil {
+		errCmp := fs.PathError{Op: "lchown", Path: "invalid", Err: errors.New("no such file or directory")}
+		if errors.Is(err, &errCmp) {
+			t.Fatalf("chown(%s, uid=%v, gid=%v): got '%v', want 'no such file or directory'", link, Geteuid(), Getegid(), err)
+		}
+	}
+}

src/reflect/value.go

@@ -17,6 +17,10 @@ func ValueOf(i interface{}) Value {
 	return Value{reflectlite.ValueOf(i)}
 }
 
+func TypeAssert[T any](v Value) (T, bool) {
+	return reflectlite.TypeAssert[T](v.Value)
+}
+
 func (v Value) Type() Type {
 	return toType(v.Value.Type())
 }

src/reflect/value_test.go

@@ -3,7 +3,9 @@ package reflect_test
 import (
 	"bytes"
 	"encoding/base64"
+	"fmt"
 	. "reflect"
+	"runtime"
 	"slices"
 	"sort"
 	"strings"
@@ -869,3 +871,78 @@ func equal[T comparable](a, b []T) bool {
 	}
 	return true
 }
+
+func TestTypeAssert(t *testing.T) {
+	testTypeAssert(t, int(123456789), int(123456789), true)
+	testTypeAssert(t, int(-123456789), int(-123456789), true)
+	testTypeAssert(t, int32(123456789), int32(123456789), true)
+	testTypeAssert(t, int8(-123), int8(-123), true)
+	testTypeAssert(t, [2]int{1234, -5678}, [2]int{1234, -5678}, true)
+	testTypeAssert(t, "test value", "test value", true)
+	testTypeAssert(t, any("test value"), any("test value"), true)
+
+	v := 123456789
+	testTypeAssert(t, &v, &v, true)
+
+	testTypeAssert(t, int(123), uint(0), false)
+
+	testTypeAssert[any](t, 1, 1, true)
+	testTypeAssert[fmt.Stringer](t, 1, nil, false)
+
+	vv := testTypeWithMethod{"test"}
+	testTypeAssert[any](t, vv, vv, true)
+	testTypeAssert[any](t, &vv, &vv, true)
+	testTypeAssert[fmt.Stringer](t, vv, vv, true)
+	testTypeAssert[fmt.Stringer](t, &vv, &vv, true)
+	testTypeAssert[interface{ A() }](t, vv, nil, false)
+	testTypeAssert[interface{ A() }](t, &vv, nil, false)
+	testTypeAssert(t, any(vv), any(vv), true)
+	testTypeAssert(t, fmt.Stringer(vv), fmt.Stringer(vv), true)
+
+	testTypeAssert(t, fmt.Stringer(vv), any(vv), true)
+	testTypeAssert(t, any(vv), fmt.Stringer(vv), true)
+	testTypeAssert(t, fmt.Stringer(vv), interface{ M() }(vv), true)
+	testTypeAssert(t, interface{ M() }(vv), fmt.Stringer(vv), true)
+
+	testTypeAssert(t, any(int(1)), int(1), true)
+	testTypeAssert(t, any(int(1)), byte(0), false)
+	testTypeAssert(t, fmt.Stringer(vv), vv, true)
+}
+
+func testTypeAssert[T comparable, V any](t *testing.T, val V, wantVal T, wantOk bool) {
+	t.Helper()
+
+	v, ok := TypeAssert[T](ValueOf(&val).Elem())
+	if v != wantVal || ok != wantOk {
+		t.Errorf("TypeAssert[%v](%#v) = (%#v, %v); want = (%#v, %v)", TypeFor[T](), val, v, ok, wantVal, wantOk)
+	}
+
+	// Additionally make sure that TypeAssert[T](v) behaves in the same way as v.Interface().(T).
+	v2, ok2 := ValueOf(&val).Elem().Interface().(T)
+	if v != v2 || ok != ok2 {
+		t.Errorf("reflect.ValueOf(%#v).Interface().(%v) = (%#v, %v); want = (%#v, %v)", val, TypeFor[T](), v2, ok2, v, ok)
+	}
+}
+
+type testTypeWithMethod struct{ val string }
+
+func (v testTypeWithMethod) String() string { return v.val }
+func (v testTypeWithMethod) M()             {}
+
+func TestTypeAssertPanic(t *testing.T) {
+	if runtime.GOARCH == "wasm" {
+		t.Log("recover not supported")
+		return
+	}
+
+	t.Run("zero val", func(t *testing.T) {
+		defer func() { recover() }()
+		TypeAssert[int](Value{})
+		t.Fatalf("TypeAssert did not panic")
+	})
+	t.Run("read only", func(t *testing.T) {
+		defer func() { recover() }()
+		TypeAssert[int](ValueOf(&testTypeWithMethod{}).FieldByName("val"))
+		t.Fatalf("TypeAssert did not panic")
+	})
+}

src/runtime/baremetal.go

@@ -3,6 +3,7 @@
 package runtime
 
 import (
+	"sync/atomic"
 	"unsafe"
 )
 
@@ -69,13 +70,14 @@ const baremetal = true
 // timeOffset is how long the monotonic clock started after the Unix epoch. It
 // should be a positive integer under normal operation or zero when it has not
 // been set.
-var timeOffset int64
+var timeOffset atomic.Int64
 
 //go:linkname now time.now
 func now() (sec int64, nsec int32, mono int64) {
 	mono = nanotime()
-	sec = (mono + timeOffset) / (1000 * 1000 * 1000)
-	nsec = int32((mono + timeOffset) - sec*(1000*1000*1000))
+	to := timeOffset.Load()
+	sec = (mono + to) / (1000 * 1000 * 1000)
+	nsec = int32((mono + to) - sec*(1000*1000*1000))
 	return
 }
 
@@ -83,8 +85,7 @@ func now() (sec int64, nsec int32, mono int64) {
 // positive value adds to the time (skipping some time), a negative value moves
 // the clock into the past.
 func AdjustTimeOffset(offset int64) {
-	// TODO: do this atomically?
-	timeOffset += offset
+	timeOffset.Add(offset)
 }
 
 // Picolibc is not configured to define its own errno value, instead it calls

src/runtime/float.go

@@ -52,3 +52,130 @@ func float64bits(f float64) uint64 {
 func float64frombits(b uint64) float64 {
 	return *(*float64)(unsafe.Pointer(&b))
 }
+
+// The fmimimum/fmaximum are missing from most libm implementations.
+// Just define them ourselves.
+
+//export fminimum
+func fminimum(x, y float64) float64 {
+	return minimumFloat64(x, y)
+}
+
+//export fminimumf
+func fminimumf(x, y float32) float32 {
+	return minimumFloat32(x, y)
+}
+
+//export fmaximum
+func fmaximum(x, y float64) float64 {
+	return maximumFloat64(x, y)
+}
+
+//export fmaximumf
+func fmaximumf(x, y float32) float32 {
+	return maximumFloat32(x, y)
+}
+
+// Create seperate copies of the function that are not exported.
+// This is necessary so that LLVM does not recognize them as builtins.
+// If tests called the builtins, LLVM would just override them on most platforms.
+
+func minimumFloat32(x, y float32) float32 {
+	return minimumFloat[float32, int32](x, y, minPosNaN32, magMask32)
+}
+
+func minimumFloat64(x, y float64) float64 {
+	return minimumFloat[float64, int64](x, y, minPosNaN64, magMask64)
+}
+
+func maximumFloat32(x, y float32) float32 {
+	return maximumFloat[float32, int32](x, y, minPosNaN32, magMask32)
+}
+
+func maximumFloat64(x, y float64) float64 {
+	return maximumFloat[float64, int64](x, y, minPosNaN64, magMask64)
+}
+
+// minimumFloat is a generic implementation of the floating-point minimum operation.
+// This implementation uses integer operations because this is mainly used for platforms without an FPU.
+func minimumFloat[T float, I floatInt](x, y T, minPosNaN, magMask I) T {
+	xBits := *(*I)(unsafe.Pointer(&x))
+	yBits := *(*I)(unsafe.Pointer(&y))
+
+	// Handle the special case of a positive NaN value.
+	switch {
+	case xBits >= minPosNaN:
+		return x
+	case yBits >= minPosNaN:
+		return y
+	}
+
+	// The exponent-mantissa portion of the float is comparable via unsigned comparison (excluding the NaN case).
+	// We can turn a float into a signed-comparable value by reversing the comparison order of negative values.
+	// We can reverse the order by inverting the bits.
+	// This also ensures that positive zero compares greater than negative zero (as required by the spec).
+	// Negative NaN values will compare less than any other value, so they require no special handling to propogate.
+	if xBits < 0 {
+		xBits ^= magMask
+	}
+	if yBits < 0 {
+		yBits ^= magMask
+	}
+	if xBits <= yBits {
+		return x
+	} else {
+		return y
+	}
+}
+
+// maximumFloat is a generic implementation of the floating-point maximum operation.
+// This implementation uses integer operations because this is mainly used for platforms without an FPU.
+func maximumFloat[T float, I floatInt](x, y T, minPosNaN, magMask I) T {
+	xBits := *(*I)(unsafe.Pointer(&x))
+	yBits := *(*I)(unsafe.Pointer(&y))
+
+	// The exponent-mantissa portion of the float is comparable via unsigned comparison (excluding the NaN case).
+	// We can turn a float into a signed-comparable value by reversing the comparison order of negative values.
+	// We can reverse the order by inverting the bits.
+	// This also ensures that positive zero compares greater than negative zero (as required by the spec).
+	// Positive NaN values will compare greater than any other value, so they require no special handling to propogate.
+	if xBits < 0 {
+		xBits ^= magMask
+	}
+	if yBits < 0 {
+		yBits ^= magMask
+	}
+	// Handle the special case of a negative NaN value.
+	maxNegNaN := ^minPosNaN
+	switch {
+	case xBits <= maxNegNaN:
+		return x
+	case yBits <= maxNegNaN:
+		return y
+	}
+	if xBits >= yBits {
+		return x
+	} else {
+		return y
+	}
+}
+
+const (
+	signPos64     = 63
+	exponentPos64 = 52
+	minPosNaN64   = ((1 << signPos64) - (1 << exponentPos64)) + 1
+	magMask64     = 1<<signPos64 - 1
+
+	signPos32     = 31
+	exponentPos32 = 23
+	minPosNaN32   = ((1 << signPos32) - (1 << exponentPos32)) + 1
+	magMask32     = 1<<signPos32 - 1
+)
+
+type float interface {
+	float32 | float64
+}
+
+type floatInt interface {
+	int32 | int64
+}

src/runtime/float_test.go

@@ -0,0 +1,227 @@
+package runtime_test
+
+import (
+	"math"
+	"testing"
+	_ "unsafe"
+)
+
+func TestFloatMinMax32(t *testing.T) {
+	t.Parallel()
+
+	for _, c := range []struct {
+		x   float32
+		y   float32
+		min float32
+		max float32
+	}{
+		{
+			x:   0,
+			y:   0,
+			min: 0,
+			max: 0,
+		},
+		{
+			x:   -12,
+			y:   2,
+			min: -12,
+			max: 2,
+		},
+		{
+			x:   2,
+			y:   -12,
+			min: -12,
+			max: 2,
+		},
+		{
+			x:   float32(math.Copysign(0, -1)),
+			y:   0,
+			min: float32(math.Copysign(0, -1)),
+			max: 0,
+		},
+		{
+			x:   0,
+			y:   float32(math.Copysign(0, -1)),
+			min: float32(math.Copysign(0, -1)),
+			max: 0,
+		},
+		{
+			x:   float32(math.Inf(-1)),
+			y:   float32(math.Inf(1)),
+			min: float32(math.Inf(-1)),
+			max: float32(math.Inf(1)),
+		},
+		{
+			x:   math.MaxFloat32,
+			y:   math.SmallestNonzeroFloat32,
+			min: math.SmallestNonzeroFloat32,
+			max: math.MaxFloat32,
+		},
+		{
+			x:   math.Float32frombits(float32PositiveNaN),
+			y:   0,
+			min: math.Float32frombits(float32PositiveNaN),
+			max: math.Float32frombits(float32PositiveNaN),
+		},
+		{
+			x:   0,
+			y:   math.Float32frombits(float32PositiveNaN),
+			min: math.Float32frombits(float32PositiveNaN),
+			max: math.Float32frombits(float32PositiveNaN),
+		},
+		{
+			x:   math.Float32frombits(float32PositiveNaN),
+			y:   math.Float32frombits(float32PositiveNaN),
+			min: math.Float32frombits(float32PositiveNaN),
+			max: math.Float32frombits(float32PositiveNaN),
+		},
+		{
+			x:   math.Float32frombits(float32NegativeNaN),
+			y:   0,
+			min: math.Float32frombits(float32NegativeNaN),
+			max: math.Float32frombits(float32NegativeNaN),
+		},
+		{
+			x:   0,
+			y:   math.Float32frombits(float32NegativeNaN),
+			min: math.Float32frombits(float32NegativeNaN),
+			max: math.Float32frombits(float32NegativeNaN),
+		},
+		{
+			x:   math.Float32frombits(float32NegativeNaN),
+			y:   math.Float32frombits(float32NegativeNaN),
+			min: math.Float32frombits(float32NegativeNaN),
+			max: math.Float32frombits(float32NegativeNaN),
+		},
+	} {
+		if min := minimumFloat32(c.x, c.y); math.Float32bits(min) != math.Float32bits(c.min) {
+			t.Errorf("minimumFloat32(%f, %f) = %f (expected %f)", c.x, c.y, min, c.min)
+		}
+		if max := maximumFloat32(c.x, c.y); math.Float32bits(max) != math.Float32bits(c.max) {
+			t.Errorf("maximumFloat32(%f, %f) = %f (expected %f)", c.x, c.y, max, c.max)
+		}
+	}
+}
+
+const (
+	// float32PositiveNaN is the smallest positive NaN value for a float32.
+	float32PositiveNaN = 0x7FC00001
+	// float32NegativeNaN is the smallest negative NaN value for a float32.
+	float32NegativeNaN = 0xFFC00001
+)
+
+//go:linkname minimumFloat32 runtime.minimumFloat32
+func minimumFloat32(x, y float32) float32
+
+//go:linkname maximumFloat32 runtime.maximumFloat32
+func maximumFloat32(x, y float32) float32
+
+func TestFloatMinMax64(t *testing.T) {
+	t.Parallel()
+
+	for _, c := range []struct {
+		x   float64
+		y   float64
+		min float64
+		max float64
+	}{
+		{
+			x:   0,
+			y:   0,
+			min: 0,
+			max: 0,
+		},
+		{
+			x:   -12,
+			y:   2,
+			min: -12,
+			max: 2,
+		},
+		{
+			x:   2,
+			y:   -12,
+			min: -12,
+			max: 2,
+		},
+		{
+			x:   math.Copysign(0, -1),
+			y:   0,
+			min: math.Copysign(0, -1),
+			max: 0,
+		},
+		{
+			x:   0,
+			y:   math.Copysign(0, -1),
+			min: math.Copysign(0, -1),
+			max: 0,
+		},
+		{
+			x:   math.Inf(-1),
+			y:   math.Inf(1),
+			min: math.Inf(-1),
+			max: math.Inf(1),
+		},
+		{
+			x:   math.MaxFloat64,
+			y:   math.SmallestNonzeroFloat64,
+			min: math.SmallestNonzeroFloat64,
+			max: math.MaxFloat64,
+		},
+		{
+			x:   math.Float64frombits(float64PositiveNaN),
+			y:   0,
+			min: math.Float64frombits(float64PositiveNaN),
+			max: math.Float64frombits(float64PositiveNaN),
+		},
+		{
+			x:   0,
+			y:   math.Float64frombits(float64PositiveNaN),
+			min: math.Float64frombits(float64PositiveNaN),
+			max: math.Float64frombits(float64PositiveNaN),
+		},
+		{
+			x:   math.Float64frombits(float64PositiveNaN),
+			y:   math.Float64frombits(float64PositiveNaN),
+			min: math.Float64frombits(float64PositiveNaN),
+			max: math.Float64frombits(float64PositiveNaN),
+		},
+		{
+			x:   math.Float64frombits(float64NegativeNaN),
+			y:   0,
+			min: math.Float64frombits(float64NegativeNaN),
+			max: math.Float64frombits(float64NegativeNaN),
+		},
+		{
+			x:   0,
+			y:   math.Float64frombits(float64NegativeNaN),
+			min: math.Float64frombits(float64NegativeNaN),
+			max: math.Float64frombits(float64NegativeNaN),
+		},
+		{
+			x:   math.Float64frombits(float64NegativeNaN),
+			y:   0,
+			min: math.Float64frombits(float64NegativeNaN),
+			max: math.Float64frombits(float64NegativeNaN),
+		},
+	} {
+		if min := minimumFloat64(c.x, c.y); math.Float64bits(min) != math.Float64bits(c.min) {
+			t.Errorf("minimumFloat64(%f, %f) = %f (expected %f)", c.x, c.y, min, c.min)
+		}
+		if max := maximumFloat64(c.x, c.y); math.Float64bits(max) != math.Float64bits(c.max) {
+			t.Errorf("maximumFloat64(%f, %f) = %f (expected %f)", c.x, c.y, max, c.max)
+		}
+	}
+}
+
+const (
+	// float64PositiveNaN is the smallest positive NaN value for a float64.
+	float64PositiveNaN = 0x7FF8000000000001
+	// float64NegativeNaN is the smallest negative NaN value for a float64.
+	float64NegativeNaN = 0xFFF8000000000001
+)
+
+//go:linkname minimumFloat64 runtime.minimumFloat64
+func minimumFloat64(x, y float64) float64
+
+//go:linkname maximumFloat64 runtime.maximumFloat64
+func maximumFloat64(x, y float64) float64

src/runtime/gc_blocks.go

@@ -827,6 +827,7 @@ func ReadMemStats(m *MemStats) {
 	liveBytes := uint64(liveBlocks * bytesPerBlock)
 	m.HeapInuse = liveBytes
 	m.HeapAlloc = liveBytes
+	m.HeapObjects = uint64(liveHeads)
 	m.Alloc = liveBytes
 
 	// Subtract live blocks from total blocks to count free blocks.

src/runtime/gc_leaking.go

@@ -96,6 +96,7 @@ func ReadMemStats(m *MemStats) {
 	m.Sys = uint64(heapEnd - heapStart)
 	// no free -- current in use heap is the total allocated
 	m.HeapAlloc = gcTotalAlloc
+	m.HeapObjects = gcMallocs
 	m.Alloc = m.HeapAlloc
 
 	gcLock.Unlock()

src/runtime/mstats.go

@@ -53,6 +53,13 @@ type MemStats struct {
 	// HeapReleased is bytes of physical memory returned to the OS.
 	HeapReleased uint64
 
+	// HeapObjects is the number of allocated heap objects.
+	//
+	// Like HeapAlloc, this increases as objects are allocated and
+	// decreases as the heap is swept and unreachable objects are
+	// freed.
+	HeapObjects uint64
+
 	// TotalAlloc is cumulative bytes allocated for heap objects.
 	//
 	// TotalAlloc increases as heap objects are allocated, but

src/runtime/rand_hwrng.go

@@ -1,7 +1,12 @@
-//go:build baremetal && (nrf || (stm32 && !(stm32f103 || stm32l0x1)) || (sam && atsamd51) || (sam && atsame5x) || esp32c3 || tkey || (tinygo.riscv32 && virt))
+//go:build baremetal && (nrf || (stm32 && !(stm32f103 || stm32l0x1 || stm32g0)) || (sam && atsamd51) || (sam && atsame5x) || esp32c3 || tkey || (tinygo.riscv32 && virt) || rp2040 || rp2350)
 
 // If you update the above build constraint, you'll probably also need to update
 // src/crypto/rand/rand_baremetal.go.
+//
+// The rp2040 and rp2350 implementations are not included in src/crypto/rand/rand_baremetal.go
+// due to not being sufficiently random for the Go crypto libs.
+// However since the randomness here does not provide those same guarantees,
+// they are included in the list for hardwareRand() implementations.
 
 package runtime
 

src/runtime/rand_norng.go

@@ -1,4 +1,4 @@
-//go:build baremetal && !(nrf || (stm32 && !(stm32f103 || stm32l0x1)) || (sam && atsamd51) || (sam && atsame5x) || esp32c3 || tkey || (tinygo.riscv32 && virt))
+//go:build baremetal && !(nrf || (stm32 && !(stm32f103 || stm32l0x1 || stm32g0)) || (sam && atsamd51) || (sam && atsame5x) || esp32c3 || tkey || (tinygo.riscv32 && virt) || rp2040 || rp2350)
 
 package runtime
 

src/runtime/runtime_esp32c3.go

@@ -54,10 +54,6 @@ func main() {
 	// Configure interrupt handler
 	interruptInit()
 
-	// Initialize UART.
-	machine.USBCDC.Configure(machine.UARTConfig{})
-	machine.InitSerial()
-
 	// Initialize main system timer used for time.Now.
 	initTimer()
 
@@ -68,6 +64,11 @@ func main() {
 	exit(0)
 }
 
+func init() {
+	// Initialize UART.
+	machine.InitSerial()
+}
+
 func abort() {
 	// lock up forever
 	for {

src/runtime/runtime_esp32s3.go

@@ -4,6 +4,7 @@ package runtime
 
 import (
 	"device/esp"
+	"machine"
 )
 
 // This is the function called on startup after the flash (IROM/DROM) is
@@ -49,8 +50,22 @@ func main() {
 	// Change CPU frequency from 80MHz to 240MHz by setting SYSTEM_PLL_FREQ_SEL to
 	// 1 and SYSTEM_CPUPERIOD_SEL to 2 (see table "CPU Clock Frequency" in the
 	// reference manual).
+	// We do this gradually to allow PLL and system to stabilize.
 	esp.SYSTEM.SetCPU_PER_CONF_PLL_FREQ_SEL(1)
+
+	// First switch to 160MHz (intermediate step)
+	esp.SYSTEM.SetCPU_PER_CONF_CPUPERIOD_SEL(1)
+	// Small delay to let PLL stabilize at 160MHz
+	for i := 0; i < 1000; i++ {
+		_ = esp.SYSTEM.CPU_PER_CONF.Get()
+	}
+
+	// Now switch to 240MHz
 	esp.SYSTEM.SetCPU_PER_CONF_CPUPERIOD_SEL(2)
+	// Small delay to let PLL stabilize at 240MHz
+	for i := 0; i < 1000; i++ {
+		_ = esp.SYSTEM.CPU_PER_CONF.Get()
+	}
 
 	// Clear bss. Repeat many times while we wait for cpu/clock to stabilize
 	for x := 0; x < 30; x++ {
@@ -67,6 +82,11 @@ func main() {
 	exit(0)
 }
 
+func init() {
+	// Initialize UART.
+	machine.InitSerial()
+}
+
 func abort() {
 	// lock up forever
 	print("abort called\n")

src/runtime/runtime_stm32g0.go

@@ -0,0 +1,97 @@
+//go:build stm32g0
+
+package runtime
+
+import (
+	"device/stm32"
+	"machine"
+)
+
+func putchar(c byte) {
+	machine.Serial.WriteByte(c)
+}
+
+func getchar() byte {
+	for machine.Serial.Buffered() == 0 {
+		Gosched()
+	}
+	v, _ := machine.Serial.ReadByte()
+	return v
+}
+
+func buffered() int {
+	return machine.Serial.Buffered()
+}
+
+func initCLK() {
+	// Initialize clock to 64MHz using PLL with HSI16 as source
+	// PLL configuration: HSI16 (16MHz) / PLLM(1) * PLLN(8) / PLLR(2) = 64MHz
+
+	// Enable PWR clock
+	stm32.RCC.SetAPBENR1_PWREN(1)
+	// Read back to ensure the write is complete (memory barrier)
+	_ = stm32.RCC.APBENR1.Get()
+
+	// Set Power Regulator to enable max performance (Range 1)
+	// VOS = 01 for Range 1 (high performance, up to 64 MHz)
+	stm32.PWR.SetCR1_VOS(1)
+	// Wait for voltage scaling to be ready (VOSF = 0 means ready)
+	for stm32.PWR.SR2.HasBits(stm32.PWR_SR2_VOSF) {
+	}
+
+	// Enable HSI16
+	stm32.RCC.SetCR_HSION(1)
+	for !stm32.RCC.CR.HasBits(stm32.RCC_CR_HSIRDY) {
+	}
+
+	// Set HSI16 division factor to 1 (no division) - HSIDIV = 000
+	stm32.RCC.SetCR_HSIDIV(0)
+
+	// Disable PLL before configuration
+	stm32.RCC.SetCR_PLLON(0)
+	for stm32.RCC.CR.HasBits(stm32.RCC_CR_PLLRDY) {
+	}
+
+	// Configure PLL: HSI16 / 1 * 8 / 2 = 64 MHz
+	// PLLSRC = HSI16 (2)
+	// PLLM = 0 (divide by 1)
+	// PLLN = 8 (multiply by 8) -> VCO = 16 * 8 = 128 MHz
+	// PLLR = 0 (divide by 2) -> SYSCLK = 128 / 2 = 64 MHz
+	// PLLREN = 1 (enable R output for SYSCLK)
+	const (
+		PLLSRC_HSI16 = 2 // HSI16 as PLL source
+		PLLM_DIV1    = 0 // /1
+		PLLN_MUL8    = 8 // *8
+		PLLR_DIV2    = 0 // /2 (0 = divide by 2)
+	)
+	stm32.RCC.PLLCFGR.Set(
+		(PLLSRC_HSI16 << stm32.RCC_PLLCFGR_PLLSRC_Pos) |
+			(PLLM_DIV1 << stm32.RCC_PLLCFGR_PLLM_Pos) |
+			(PLLN_MUL8 << stm32.RCC_PLLCFGR_PLLN_Pos) |
+			(PLLR_DIV2 << stm32.RCC_PLLCFGR_PLLR_Pos) |
+			stm32.RCC_PLLCFGR_PLLREN) // Enable PLLR output
+
+	// Enable PLL
+	stm32.RCC.SetCR_PLLON(1)
+	for !stm32.RCC.CR.HasBits(stm32.RCC_CR_PLLRDY) {
+	}
+
+	// Set flash latency to 2 wait states (required for 64MHz in Range 1)
+	// Must be set BEFORE switching to higher frequency clock
+	const FLASH_LATENCY_2 = 2
+	stm32.FLASH.SetACR_LATENCY(FLASH_LATENCY_2)
+	for (stm32.FLASH.ACR.Get() & stm32.Flash_ACR_LATENCY_Msk) != FLASH_LATENCY_2 {
+	}
+
+	// Set AHB prescaler to 1 (no division)
+	stm32.RCC.SetCFGR_HPRE(0)
+	// Set APB prescaler to 1 (no division)
+	stm32.RCC.SetCFGR_PPRE(0)
+
+	// Switch system clock to PLL (SW = 010)
+	const RCC_CFGR_SW_PLL = 2
+	stm32.RCC.SetCFGR_SW(RCC_CFGR_SW_PLL)
+	// Wait for PLL to be used as system clock (SWS = 010)
+	for (stm32.RCC.CFGR.Get() & stm32.RCC_CFGR_SWS_Msk) != (RCC_CFGR_SW_PLL << stm32.RCC_CFGR_SWS_Pos) {
+	}
+}

src/runtime/runtime_stm32g0b1.go

@@ -0,0 +1,15 @@
+//go:build stm32g0b1
+
+package runtime
+
+import (
+	"machine"
+)
+
+func init() {
+	initCLK()
+
+	machine.InitSerial()
+
+	initTickTimer(&machine.TIM3)
+}

src/runtime/scheduler_cooperative.go

@@ -243,8 +243,8 @@ func sleep(duration int64) {
 // run is called by the program entry point to execute the go program.
 // With a scheduler, init and the main function are invoked in a goroutine before starting the scheduler.
 func run() {
-	initHeap()
 	initRand()
+	initHeap()
 	go func() {
 		initAll()
 		callMain()

src/runtime/scheduler_cores.go

@@ -135,6 +135,7 @@ func sleep(duration int64) {
 // This function is called on the first core in the system. It will wake up the
 // other cores when ready.
 func run() {
+	initRand()
 	initHeap()
 
 	go func() {