A MultiArrayList stores a list of a struct or tagged union type.
Instead of storing a single list of items, MultiArrayList
stores separate lists for each field of the struct or
lists of tags and bare unions.
This allows for memory savings if the struct or union has padding,
and also improves cache usage if only some fields or just tags
are needed for a computation. The primary API for accessing fields is
the slice() function, which computes the start pointers
for the array of each field. From the slice you can call
.items(.<field_name>) to obtain a slice of field values.
For unions you can call .items(.tags) or .items(.data).
pub fn MultiArrayList(comptime T: type) typepub fn MultiArrayList(comptime T: type) type {
return struct {
/// This pointer is always aligned to the boundary `sizes.big_align`; this is not specified
/// in the type to avoid `MultiArrayList(T)` depending on the alignment of `T` because this
/// can lead to dependency loops. See `allocatedBytes` which `@alignCast`s this pointer to
/// the correct type.
bytes: [*]u8 = undefined,
len: usize = 0,
capacity: usize = 0,
pub const empty: Self = .{
.bytes = undefined,
.len = 0,
.capacity = 0,
};
/// Initialize with capacity to hold exactly `num` elements.
/// Deinitialize with `deinit` or `toOwnedSlice`.
pub fn initCapacity(gpa: Allocator, num: usize) Allocator.Error!Self {
var self: Self = .empty;
try self.setCapacity(gpa, num);
return self;
}
const Elem = switch (@typeInfo(T)) {
.@"struct" => T,
.@"union" => |u| struct {
pub const Bare = std.meta.BareUnion(T);
pub const Tag =
u.tag_type orelse @compileError("MultiArrayList does not support untagged unions");
tags: Tag,
data: Bare,
pub fn fromT(outer: T) @This() {
const tag = meta.activeTag(outer);
return .{
.tags = tag,
.data = switch (tag) {
inline else => |t| @unionInit(Bare, @tagName(t), @field(outer, @tagName(t))),
},
};
}
pub fn toT(tag: Tag, bare: Bare) T {
return switch (tag) {
inline else => |t| @unionInit(T, @tagName(t), @field(bare, @tagName(t))),
};
}
},
else => @compileError("MultiArrayList only supports structs and tagged unions"),
};
pub const Field = meta.FieldEnum(Elem);
/// A MultiArrayList.Slice contains cached start pointers for each field in the list.
/// These pointers are not normally stored to reduce the size of the list in memory.
/// If you are accessing multiple fields, call slice() first to compute the pointers,
/// and then get the field arrays from the slice.
pub const Slice = struct {
/// This array is indexed by the field index which can be obtained
/// by using @intFromEnum() on the Field enum
ptrs: [field_names.len][*]u8,
len: usize,
capacity: usize,
pub const empty: Slice = .{
.ptrs = undefined,
.len = 0,
.capacity = 0,
};
pub fn items(self: Slice, comptime field: Field) []FieldType(field) {
const F = FieldType(field);
if (self.capacity == 0) {
return &[_]F{};
}
const byte_ptr = self.ptrs[@backingInt(field)];
const casted_ptr: [*]F = if (@sizeOf(F) == 0)
undefined
else
@ptrCast(@alignCast(byte_ptr));
return casted_ptr[0..self.len];
}
pub fn set(self: *Slice, index: usize, elem: T) void {
const e = switch (@typeInfo(T)) {
.@"struct" => elem,
.@"union" => Elem.fromT(elem),
else => unreachable,
};
inline for (field_names, 0..) |field_name, i| {
self.items(@as(Field, @fromBackingInt(@intCast(i))))[index] = @field(e, field_name);
}
}
pub fn get(self: Slice, index: usize) T {
var result: Elem = undefined;
inline for (field_names, 0..) |field_name, i| {
@field(result, field_name) = self.items(@as(Field, @fromBackingInt(@intCast(i))))[index];
}
return switch (@typeInfo(T)) {
.@"struct" => result,
.@"union" => Elem.toT(result.tags, result.data),
else => unreachable,
};
}
pub fn swap(self: Slice, a: usize, b: usize) void {
inline for (@typeInfo(Field).@"enum".field_names) |field_name| {
const its = self.items(@field(Field, field_name));
std.mem.swap(@FieldType(T, field_name), &its[a], &its[b]);
}
}
pub fn toMultiArrayList(self: Slice) Self {
if (self.ptrs.len == 0 or self.capacity == 0) {
return .{};
}
return .{
.bytes = self.ptrs[sizes.fields[0]],
.len = self.len,
.capacity = self.capacity,
};
}
pub fn deinit(self: *Slice, gpa: Allocator) void {
var other = self.toMultiArrayList();
other.deinit(gpa);
self.* = undefined;
}
/// Returns a `Slice` representing a range of elements in `s`, analagous to `arr[off..len]`.
/// It is illegal to call `deinit` or `toMultiArrayList` on the returned `Slice`.
/// Asserts that `off + len <= s.len`.
pub fn subslice(s: Slice, off: usize, len: usize) Slice {
assert(off + len <= s.len);
var ptrs: [field_names.len][*]u8 = undefined;
inline for (s.ptrs, &ptrs, field_types) |in, *out, field_type| {
out.* = in + (off * @sizeOf(field_type));
}
return .{
.ptrs = ptrs,
.len = len,
.capacity = len,
};
}
/// This function is used in the debugger pretty formatters in tools/ to fetch the
/// child field order and entry type to facilitate fancy debug printing for this type.
fn dbHelper(self: *Slice, child: *Elem, field: *Field, entry: *Entry) void {
_ = self;
_ = child;
_ = field;
_ = entry;
}
};
const Self = @This();
const field_names = @typeInfo(Elem).@"struct".field_names;
const field_types = @typeInfo(Elem).@"struct".field_types;
const field_attrs = @typeInfo(Elem).@"struct".field_attrs;
/// `sizes.bytes` is an array of @sizeOf each T field. Sorted by alignment, descending.
/// `sizes.fields` is an array mapping from `sizes.bytes` array index to field index.
/// `sizes.big_align` is the overall alignment of the allocation, which equals the maximum field alignment.
const sizes = blk: {
const Data = struct {
size: usize,
size_index: usize,
alignment: usize,
};
var data: [field_names.len]Data = undefined;
var big_align: usize = 1;
for (field_types, field_attrs, 0..) |f_type, f_attrs, i| {
data[i] = .{
.size = @sizeOf(f_type),
.size_index = i,
.alignment = f_attrs.@"align" orelse @alignOf(f_type),
};
big_align = @max(big_align, data[i].alignment);
}
const Sort = struct {
fn lessThan(context: void, lhs: Data, rhs: Data) bool {
_ = context;
return lhs.alignment > rhs.alignment;
}
};
@setEvalBranchQuota(3 * field_names.len * std.math.log2(field_names.len));
mem.sort(Data, &data, {}, Sort.lessThan);
var sizes_bytes: [field_names.len]usize = undefined;
var field_indexes: [field_names.len]usize = undefined;
for (data, 0..) |elem, i| {
sizes_bytes[i] = elem.size;
field_indexes[i] = elem.size_index;
}
break :blk .{
.bytes = sizes_bytes,
.fields = field_indexes,
.big_align = mem.Alignment.fromByteUnits(big_align),
};
};
/// Release all allocated memory.
pub fn deinit(self: *Self, gpa: Allocator) void {
gpa.free(self.allocatedBytes());
self.* = undefined;
}
/// The caller owns the returned memory. Empties this MultiArrayList.
pub fn toOwnedSlice(self: *Self) Slice {
const result = self.slice();
self.* = .{};
return result;
}
/// Compute pointers to the start of each field of the array.
/// If you need to access multiple fields, calling this may
/// be more efficient than calling `items()` multiple times.
pub fn slice(self: Self) Slice {
var result: Slice = .{
.ptrs = undefined,
.len = self.len,
.capacity = self.capacity,
};
var ptr: [*]u8 = self.bytes;
for (sizes.bytes, sizes.fields) |field_size, i| {
result.ptrs[i] = ptr;
ptr += field_size * self.capacity;
}
return result;
}
/// Get the slice of values for a specified field.
/// If you need multiple fields, consider calling slice()
/// instead.
pub fn items(self: Self, comptime field: Field) []FieldType(field) {
return self.slice().items(field);
}
/// Overwrite one array element with new data.
pub fn set(self: *Self, index: usize, elem: T) void {
var slices = self.slice();
slices.set(index, elem);
}
/// Obtain all the data for one array element.
pub fn get(self: Self, index: usize) T {
return self.slice().get(index);
}
pub fn swap(self: Self, a: usize, b: usize) void {
return self.slice().swap(a, b);
}
/// Extend the list by 1 element.
///
/// Allocates more memory as necessary.
pub fn append(self: *Self, gpa: Allocator, elem: T) Allocator.Error!void {
try self.ensureUnusedCapacity(gpa, 1);
self.appendAssumeCapacity(elem);
}
/// Extend the list by 1 element.
///
/// Asserts that capacity is sufficient to hold an additional item.
pub fn appendAssumeCapacity(self: *Self, elem: T) void {
assert(self.len < self.capacity);
self.len += 1;
self.set(self.len - 1, elem);
}
/// Extend the list by 1 element.
///
/// If capacity is not sufficient to hold an additional
/// item, returns `error.OutOfMemory`.
pub fn appendBounded(self: *Self, elem: T) error{OutOfMemory}!void {
if (self.capacity - self.len < 1) return error.OutOfMemory;
return appendAssumeCapacity(self, elem);
}
/// Extend the list by 1 element, returning the newly reserved
/// index with uninitialized data.
///
/// Allocates more memory as necessary.
pub fn addOne(self: *Self, gpa: Allocator) Allocator.Error!usize {
try self.ensureUnusedCapacity(gpa, 1);
return self.addOneAssumeCapacity();
}
/// Extend the list by 1 element, returning the newly reserved
/// index with uninitialized data.
///
/// Asserts that capacity is sufficient to hold an additional item.
pub fn addOneAssumeCapacity(self: *Self) usize {
assert(self.len < self.capacity);
const index = self.len;
self.len += 1;
return index;
}
/// Extend the list by 1 element, returning the newly reserved
/// index with uninitialized data.
///
/// If capacity is not sufficient to hold an additional
/// item, returns `error.OutOfMemory`.
pub fn addOneBounded(self: *Self) error{OutOfMemory}!usize {
if (self.capacity - self.len < 1) return error.OutOfMemory;
return addOneAssumeCapacity(self);
}
/// Remove and return the last element from the list, or return `null` if list is empty.
/// Invalidates pointers to fields of the removed element.
pub fn pop(self: *Self) ?T {
if (self.len == 0) return null;
const val = self.get(self.len - 1);
self.len -= 1;
return val;
}
/// Inserts an item into the list. Shifts all elements
/// after and including the specified index back by one and
/// sets the given index to the specified element.
///
/// Allocates more memory as necessary.
pub fn insert(self: *Self, gpa: Allocator, index: usize, elem: T) !void {
try self.ensureUnusedCapacity(gpa, 1);
self.insertAssumeCapacity(index, elem);
}
/// Inserts an item into the list. Shifts all elements
/// after and including the specified index back by one and
/// sets the given index to the specified element.
///
/// Asserts that capacity is sufficient to hold an additional item.
pub fn insertAssumeCapacity(self: *Self, index: usize, elem: T) void {
assert(self.len < self.capacity);
assert(index <= self.len);
self.len += 1;
const entry = switch (@typeInfo(T)) {
.@"struct" => elem,
.@"union" => Elem.fromT(elem),
else => unreachable,
};
const slices = self.slice();
inline for (field_names, 0..) |field_name, field_index| {
const field_slice = slices.items(@as(Field, @fromBackingInt(@intCast(field_index))));
var i: usize = self.len - 1;
while (i > index) : (i -= 1) {
field_slice[i] = field_slice[i - 1];
}
field_slice[index] = @field(entry, field_name);
}
}
/// Inserts an item into the list. Shifts all elements
/// after and including the specified index back by one and
/// sets the given index to the specified element.
///
/// If capacity is not sufficient to hold an additional
/// item, returns `error.OutOfMemory`.
pub fn insertBounded(self: *Self, index: usize, elem: T) error{OutOfMemory}!void {
if (self.capacity - self.len < 1) return error.OutOfMemory;
return insertAssumeCapacity(self, index, elem);
}
/// Remove the specified item from the list, swapping the last
/// item in the list into its position. Fast, but does not
/// retain list ordering.
pub fn swapRemove(self: *Self, index: usize) void {
const slices = self.slice();
inline for (field_names, 0..) |_, i| {
const field_slice = slices.items(@as(Field, @fromBackingInt(@intCast(i))));
field_slice[index] = field_slice[self.len - 1];
field_slice[self.len - 1] = undefined;
}
self.len -= 1;
}
/// Remove the specified item from the list, shifting items
/// after it to preserve order.
pub fn orderedRemove(self: *Self, index: usize) void {
const slices = self.slice();
inline for (field_names, 0..) |_, field_index| {
const field_slice = slices.items(@as(Field, @fromBackingInt(@intCast(field_index))));
var i = index;
while (i < self.len - 1) : (i += 1) {
field_slice[i] = field_slice[i + 1];
}
field_slice[i] = undefined;
}
self.len -= 1;
}
/// Remove the elements indexed by `sorted_indexes`. The indexes to be
/// removed correspond to the array list before deletion.
///
/// Asserts:
/// * Each index to be removed is in bounds.
/// * The indexes to be removed are sorted ascending.
///
/// Duplicates in `sorted_indexes` are allowed.
///
/// This operation is O(N).
///
/// Invalidates element pointers beyond the first deleted index.
pub fn orderedRemoveMany(self: *Self, sorted_indexes: []const usize) void {
if (sorted_indexes.len == 0) return;
const slices = self.slice();
var shift: usize = 1;
for (sorted_indexes[0 .. sorted_indexes.len - 1], sorted_indexes[1..]) |removed, end| {
if (removed == end) continue; // allows duplicates in `sorted_indexes`
const start = removed + 1;
const len = end - start; // safety checks `sorted_indexes` are sorted
inline for (field_names, 0..) |_, field_index| {
const field_slice = slices.items(@fromBackingInt(@intCast(field_index)));
@memmove(field_slice[start - shift ..][0..len], field_slice[start..][0..len]); // safety checks initial `sorted_indexes` are in range
}
shift += 1;
}
const start = sorted_indexes[sorted_indexes.len - 1] + 1;
const end = self.len;
const len = end - start; // safety checks final `sorted_indexes` are in range
inline for (field_names, 0..) |_, field_index| {
const field_slice = slices.items(@fromBackingInt(@intCast(field_index)));
@memmove(field_slice[start - shift ..][0..len], field_slice[start..][0..len]);
}
self.len = end - shift;
}
/// Adjust the list's length to `new_len`.
/// Does not initialize added items, if any.
pub fn resize(self: *Self, gpa: Allocator, new_len: usize) Allocator.Error!void {
try self.ensureTotalCapacity(gpa, new_len);
self.len = new_len;
}
/// Attempt to reduce allocated capacity to `new_len`.
/// If `new_len` is greater than zero, this may fail to reduce the capacity,
/// but the data remains intact and the length is updated to new_len.
pub fn shrinkAndFree(self: *Self, gpa: Allocator, new_len: usize) void {
if (new_len == 0) return clearAndFree(self, gpa);
assert(new_len <= self.capacity);
assert(new_len <= self.len);
const other_bytes = gpa.alignedAlloc(u8, sizes.big_align, capacityInBytes(new_len)) catch {
const self_slice = self.slice();
inline for (field_types, 0..) |field_type, i| {
if (@sizeOf(field_type) != 0) {
const field = @as(Field, @fromBackingInt(@intCast(i)));
const dest_slice = self_slice.items(field)[new_len..];
// We use memset here for more efficient codegen in safety-checked,
// valgrind-enabled builds. Otherwise the valgrind client request
// will be repeated for every element.
@memset(dest_slice, undefined);
}
}
self.len = new_len;
return;
};
var other = Self{
.bytes = other_bytes.ptr,
.capacity = new_len,
.len = new_len,
};
self.len = new_len;
const self_slice = self.slice();
const other_slice = other.slice();
inline for (field_types, 0..) |field_type, i| {
if (@sizeOf(field_type) != 0) {
const field = @as(Field, @fromBackingInt(@intCast(i)));
@memcpy(other_slice.items(field), self_slice.items(field));
}
}
gpa.free(self.allocatedBytes());
self.* = other;
}
pub fn clearAndFree(self: *Self, gpa: Allocator) void {
gpa.free(self.allocatedBytes());
self.* = .{};
}
/// Reduce length to `new_len`.
/// Invalidates pointers to elements `items[new_len..]`.
/// Keeps capacity the same.
pub fn shrinkRetainingCapacity(self: *Self, new_len: usize) void {
self.len = new_len;
}
/// Invalidates all element pointers.
pub fn clearRetainingCapacity(self: *Self) void {
self.len = 0;
}
/// Modify the array so that it can hold at least `new_capacity` items.
/// Implements super-linear growth to achieve amortized O(1) append operations.
/// Invalidates element pointers if additional memory is needed.
pub fn ensureTotalCapacity(self: *Self, gpa: Allocator, new_capacity: usize) Allocator.Error!void {
if (self.capacity >= new_capacity) return;
return self.setCapacity(gpa, growCapacity(new_capacity));
}
const init_capacity: comptime_int = init: {
var max: comptime_int = 1;
for (field_types) |field_type| max = @max(max, @sizeOf(field_type));
break :init @max(1, std.atomic.cache_line / max);
};
/// Given a lower bound of required memory capacity, returns a larger value
/// with super-linear growth.
pub fn growCapacity(minimum: usize) usize {
return minimum +| (minimum / 2 + init_capacity);
}
/// Modify the array so that it can hold at least `additional_count` **more** items.
/// Invalidates pointers if additional memory is needed.
pub fn ensureUnusedCapacity(self: *Self, gpa: Allocator, additional_count: usize) Allocator.Error!void {
return self.ensureTotalCapacity(gpa, self.len + additional_count);
}
/// Modify the array so that it can hold exactly `new_capacity` items.
/// Invalidates pointers if additional memory is needed.
/// `new_capacity` must be greater or equal to `len`.
pub fn setCapacity(self: *Self, gpa: Allocator, new_capacity: usize) Allocator.Error!void {
assert(new_capacity >= self.len);
const new_bytes = try gpa.alignedAlloc(u8, sizes.big_align, capacityInBytes(new_capacity));
if (self.len == 0) {
gpa.free(self.allocatedBytes());
self.bytes = new_bytes.ptr;
self.capacity = new_capacity;
return;
}
var other = Self{
.bytes = new_bytes.ptr,
.capacity = new_capacity,
.len = self.len,
};
const self_slice = self.slice();
const other_slice = other.slice();
inline for (field_types, 0..) |field_type, i| {
if (@sizeOf(field_type) != 0) {
const field = @as(Field, @fromBackingInt(@intCast(i)));
@memcpy(other_slice.items(field), self_slice.items(field));
}
}
gpa.free(self.allocatedBytes());
self.* = other;
}
/// Create a copy of this list with a new backing store,
/// using the specified allocator.
pub fn clone(self: Self, gpa: Allocator) Allocator.Error!Self {
var result = Self{};
errdefer result.deinit(gpa);
try result.ensureTotalCapacity(gpa, self.len);
result.len = self.len;
const self_slice = self.slice();
const result_slice = result.slice();
inline for (field_types, 0..) |field_type, i| {
if (@sizeOf(field_type) != 0) {
const field = @as(Field, @fromBackingInt(@intCast(i)));
@memcpy(result_slice.items(field), self_slice.items(field));
}
}
return result;
}
/// `ctx` has the following method:
/// `fn lessThan(ctx: @TypeOf(ctx), a_index: usize, b_index: usize) bool`
fn sortInternal(self: Self, a: usize, b: usize, ctx: anytype, comptime mode: std.sort.Mode) void {
const sort_context: struct {
sub_ctx: @TypeOf(ctx),
slice: Slice,
pub fn swap(sc: @This(), a_index: usize, b_index: usize) void {
inline for (field_types, 0..) |field_type, i| {
if (@sizeOf(field_type) != 0) {
const field: Field = @fromBackingInt(@intCast(i));
const ptr = sc.slice.items(field);
mem.swap(field_type, &ptr[a_index], &ptr[b_index]);
}
}
}
pub fn lessThan(sc: @This(), a_index: usize, b_index: usize) bool {
return sc.sub_ctx.lessThan(a_index, b_index);
}
} = .{
.sub_ctx = ctx,
.slice = self.slice(),
};
switch (mode) {
.stable => mem.sortContext(a, b, sort_context),
.unstable => mem.sortUnstableContext(a, b, sort_context),
}
}
/// This function guarantees a stable sort, i.e the relative order of equal elements is preserved during sorting.
/// Read more about stable sorting here: https://en.wikipedia.org/wiki/Sorting_algorithm#Stability
/// If this guarantee does not matter, `sortUnstable` might be a faster alternative.
/// `ctx` has the following method:
/// `fn lessThan(ctx: @TypeOf(ctx), a_index: usize, b_index: usize) bool`
pub fn sort(self: Self, ctx: anytype) void {
self.sortInternal(0, self.len, ctx, .stable);
}
/// Sorts only the subsection of items between indices `a` and `b` (excluding `b`)
/// This function guarantees a stable sort, i.e the relative order of equal elements is preserved during sorting.
/// Read more about stable sorting here: https://en.wikipedia.org/wiki/Sorting_algorithm#Stability
/// If this guarantee does not matter, `sortSpanUnstable` might be a faster alternative.
/// `ctx` has the following method:
/// `fn lessThan(ctx: @TypeOf(ctx), a_index: usize, b_index: usize) bool`
pub fn sortSpan(self: Self, a: usize, b: usize, ctx: anytype) void {
self.sortInternal(a, b, ctx, .stable);
}
/// This function does NOT guarantee a stable sort, i.e the relative order of equal elements may change during sorting.
/// Due to the weaker guarantees of this function, this may be faster than the stable `sort` method.
/// Read more about stable sorting here: https://en.wikipedia.org/wiki/Sorting_algorithm#Stability
/// `ctx` has the following method:
/// `fn lessThan(ctx: @TypeOf(ctx), a_index: usize, b_index: usize) bool`
pub fn sortUnstable(self: Self, ctx: anytype) void {
self.sortInternal(0, self.len, ctx, .unstable);
}
/// Sorts only the subsection of items between indices `a` and `b` (excluding `b`)
/// This function does NOT guarantee a stable sort, i.e the relative order of equal elements may change during sorting.
/// Due to the weaker guarantees of this function, this may be faster than the stable `sortSpan` method.
/// Read more about stable sorting here: https://en.wikipedia.org/wiki/Sorting_algorithm#Stability
/// `ctx` has the following method:
/// `fn lessThan(ctx: @TypeOf(ctx), a_index: usize, b_index: usize) bool`
pub fn sortSpanUnstable(self: Self, a: usize, b: usize, ctx: anytype) void {
self.sortInternal(a, b, ctx, .unstable);
}
pub fn capacityInBytes(capacity: usize) usize {
comptime var elem_bytes: usize = 0;
inline for (sizes.bytes) |size| elem_bytes += size;
return elem_bytes * capacity;
}
fn allocatedBytes(self: Self) []align(sizes.big_align.toByteUnits()) u8 {
return @alignCast(self.bytes[0..capacityInBytes(self.capacity)]);
}
fn FieldType(comptime field: Field) type {
return @FieldType(Elem, @tagName(field));
}
const Entry = entry: {
var entry_field_names: [field_names.len][]const u8 = undefined;
var entry_field_types: [field_names.len]type = undefined;
var entry_field_attrs: [field_names.len]std.builtin.Type.Struct.FieldAttributes = undefined;
for (sizes.fields, &entry_field_names, &entry_field_types, &entry_field_attrs) |i, *name, *Type, *attrs| {
name.* = field_names[i] ++ "_ptr";
Type.* = *field_types[i];
attrs.* = .{
.@"comptime" = field_attrs[i].@"comptime",
.@"align" = field_attrs[i].@"align",
};
}
break :entry @Struct(.@"extern", null, &entry_field_names, &entry_field_types, &entry_field_attrs);
};
/// This function is used in the debugger pretty formatters in tools/ to fetch the
/// child field order and entry type to facilitate fancy debug printing for this type.
fn dbHelper(self: *Self, child: *Elem, field: *Field, entry: *Entry) void {
_ = self;
_ = child;
_ = field;
_ = entry;
}
comptime {
if (builtin.zig_backend == .stage2_llvm and !builtin.strip_debug_info) {
_ = &dbHelper;
_ = &Slice.dbHelper;
}
}
};
}