compiler: simplify createObjectLayout

This simplifies the process of constructing and encoding layout bitmaps.
Instead of creating big integers and merging them, we can create a pre-sized bitmap and set positions within it.

This also changes the encoding logic to allow larger layouts to be encoded inline.
We would previously not encode a layout inline unless the size was less than the width of the data field.
This is overly conservative.
A layout can be encoded inline as long as:
1. The size fits within the size field.
2. All set bits in the bitmap fit into the data field.

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/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
+	// Create the pointer bitmap.
-	// information at all.
 	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
+	if bitmapLen < pointerBits {
-	switch pointerBits {
+		rawMask := binary.LittleEndian.Uint64(bitmap[0:8])
-	case 16:
+		layout := rawMask*pointerBits + bitmapLen
-		sizeFieldBits = 4
+		layout <<= 1
-	case 32:
+		layout |= 1
-		sizeFieldBits = 5
-	case 64:
-		sizeFieldBits = 6
-	default:
-		panic("unknown pointer size")
-	}
-	layoutFieldBits := pointerBits - 1 - sizeFieldBits
 
-	// Try to emit the value as an inline integer. This is possible in most
+		// Check if the layout fits.
-	// cases.
+		layout &= 1<<pointerBits - 1
-	if objectSizeWords < layoutFieldBits {
+		if (layout>>1)/pointerBits == rawMask {
-		// If it can be stored directly in the pointer value, do so.
+			// No set bits were shifted off.
-		// The runtime knows that if the least significant bit of the pointer is
+			return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
-		// 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)
 	}
 
 	// 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)
+	bitmapByteValues := make([]llvm.Value, bitmapBytes)
-	bitmap.FillBytes(bitmapBytes)
+	i8 := c.ctx.Int8Type()
-	reverseBytes(bitmapBytes) // big-endian to little-endian
+	for i, b := range bitmap {
-	var bitmapByteValues []llvm.Value
+		bitmapByteValues[i] = llvm.ConstInt(i8, uint64(b), false)
-	for _, b := range bitmapBytes {
-		bitmapByteValues = append(bitmapByteValues, llvm.ConstInt(c.ctx.Int8Type(), uint64(b), false))
 	}
 	initializer := c.ctx.ConstStruct([]llvm.Value{
-		llvm.ConstInt(c.uintptrType, objectSizeWords, false),
+		llvm.ConstInt(c.uintptrType, bitmapLen, false),
-		llvm.ConstArray(c.ctx.Int8Type(), bitmapByteValues),
+		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
+// buildPointerBitmap 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.
+// bitmap at the word offset that contains a pointer. This scan is recursive.
-func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *big.Int {
+func (c *compilerContext) buildPointerBitmap(
-	alignment := c.targetData.PrefTypeAlignment(c.dataPtrType)
+	dst []byte,
-	switch typ.TypeKind() {
+	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)
+		// Recurse over struct elements.
-		for i, subtyp := range typ.StructElementTypes() {
+		for i, et := range t.StructElementTypes() {
-			subptrs := c.getPointerBitmap(subtyp, pos)
+			eo := c.targetData.ElementOffset(t, i)
-			if subptrs.BitLen() == 0 {
+			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)
+			c.buildPointerBitmap(
-			if offset%uint64(alignment) != 0 {
+				dst,
-				// This error will let the compilation fail, but by continuing
+				ptrAlign,
-				// the error can still easily be shown.
+				pos,
-				c.addError(pos, "internal error: allocated struct contains unaligned pointer")
+				et,
-				continue
+				offset+(eo/ptrAlign),
-			}
+			)
-			subptrs.Lsh(subptrs, uint(offset)/uint(alignment))
-			ptrs.Or(ptrs, subptrs)
 		}
-		return ptrs
+
 	case llvm.ArrayTypeKind:
-		subtyp := typ.ElementType()
+		// Recurse over array elements.
-		subptrs := c.getPointerBitmap(subtyp, pos)
+		len := t.ArrayLength()
-		ptrs := big.NewInt(0)
+		if len <= 0 {
-		if subptrs.BitLen() == 0 {
+			return
-			return ptrs
 		}
-		elementSize := c.targetData.TypeAllocSize(subtyp)
+		et := t.ElementType()
-		if elementSize%uint64(alignment) != 0 {
+		elementSize := c.targetData.TypeAllocSize(et)
-			// This error will let the compilation fail (but continues so that
+		if elementSize%ptrAlign != 0 {
-			// other errors can be shown).
+			if typeHasPointers(et) {
-			c.addError(pos, "internal error: allocated array contains unaligned pointer")
+				// This error will let the compilation fail (but continues so that
-			return ptrs
+				// other errors can be shown).
+				c.addError(pos, "internal error: allocated array contains unaligned pointer")
+			}
+			return
 		}
-		for i := 0; i < typ.ArrayLength(); i++ {
+		elementSize /= ptrAlign
-			ptrs.Lsh(ptrs, uint(elementSize)/uint(alignment))
+		for i := 0; i < len; i++ {
-			ptrs.Or(ptrs, subptrs)
+			c.buildPointerBitmap(
+				dst,
+				ptrAlign,
+				pos,
+				et,
+				offset+uint64(i)*elementSize,
+			)
 		}
-		return ptrs
+
 	default:
 		// Should not happen.
 		panic("unknown LLVM type")

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:62-0100000000000020" = 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:62-0001" = 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:62-0100000000000000" = 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
 }