Sema: automatically optimize order of struct fields

lib/std/os/linux/io_uring.zig

@@ -954,7 +954,7 @@ pub const IO_Uring = struct {
     pub fn register_files_update(self: *IO_Uring, offset: u32, fds: []const os.fd_t) !void {
         assert(self.fd >= 0);
 
-        const FilesUpdate = struct {
+        const FilesUpdate = extern struct {
             offset: u32,
             resv: u32,
             fds: u64 align(8),

src/Module.zig

@@ -941,7 +941,8 @@ pub const Struct = struct {
     owner_decl: Decl.Index,
     /// Index of the struct_decl ZIR instruction.
     zir_index: Zir.Inst.Index,
-
+    /// Indexes into `fields` sorted to be most memory efficient.
+    optimized_order: ?[*]u32 = null,
     layout: std.builtin.Type.ContainerLayout,
     /// If the layout is not packed, this is the noreturn type.
     /// If the layout is packed, this is the backing integer type of the packed struct.
@@ -1023,6 +1024,10 @@ pub const Struct = struct {
         }
     };
 
+    /// Used in `optimized_order` to indicate field that is not present in the
+    /// runtime version of the struct.
+    pub const omitted_field = std.math.maxInt(u32);
+
     pub fn getFullyQualifiedName(s: *Struct, mod: *Module) ![:0]u8 {
         return mod.declPtr(s.owner_decl).getFullyQualifiedName(mod);
     }
@@ -1098,6 +1103,39 @@ pub const Struct = struct {
         }
         unreachable; // index out of bounds
     }
+
+    pub const RuntimeFieldIterator = struct {
+        struct_obj: *const Struct,
+        index: u32 = 0,
+
+        pub const FieldAndIndex = struct {
+            field: Field,
+            index: u32,
+        };
+
+        pub fn next(it: *RuntimeFieldIterator) ?FieldAndIndex {
+            while (true) {
+                var i = it.index;
+                it.index += 1;
+                if (it.struct_obj.fields.count() <= i)
+                    return null;
+
+                if (it.struct_obj.optimized_order) |some| {
+                    i = some[i];
+                    if (i == Module.Struct.omitted_field) return null;
+                }
+                const field = it.struct_obj.fields.values()[i];
+
+                if (!field.is_comptime and field.ty.hasRuntimeBits()) {
+                    return FieldAndIndex{ .index = i, .field = field };
+                }
+            }
+        }
+    };
+
+    pub fn runtimeFieldIterator(s: *const Struct) RuntimeFieldIterator {
+        return .{ .struct_obj = s };
+    }
 };
 
 /// Represents the data that an enum declaration provides, when the fields
@@ -6481,6 +6519,7 @@ pub const Feature = enum {
     error_return_trace,
     is_named_enum_value,
     error_set_has_value,
+    field_reordering,
 };
 
 pub fn backendSupportsFeature(mod: Module, feature: Feature) bool {
@@ -6493,5 +6532,6 @@ pub fn backendSupportsFeature(mod: Module, feature: Feature) bool {
         .error_return_trace => mod.comp.bin_file.options.use_llvm,
         .is_named_enum_value => mod.comp.bin_file.options.use_llvm,
         .error_set_has_value => mod.comp.bin_file.options.use_llvm,
+        .field_reordering => mod.comp.bin_file.options.use_llvm,
     };
 }

src/Sema.zig

@@ -29919,6 +29919,42 @@ fn resolveStructLayout(sema: *Sema, ty: Type) CompileError!void {
             );
             return sema.failWithOwnedErrorMsg(msg);
         }
+
+        if (struct_obj.layout == .Auto and sema.mod.backendSupportsFeature(.field_reordering)) {
+            const optimized_order = blk: {
+                const decl = sema.mod.declPtr(struct_obj.owner_decl);
+                var decl_arena = decl.value_arena.?.promote(sema.mod.gpa);
+                defer decl.value_arena.?.* = decl_arena.state;
+                const decl_arena_allocator = decl_arena.allocator();
+
+                break :blk try decl_arena_allocator.alloc(u32, struct_obj.fields.count());
+            };
+
+            for (struct_obj.fields.values()) |field, i| {
+                optimized_order[i] = if (field.ty.hasRuntimeBits())
+                    @intCast(u32, i)
+                else
+                    Module.Struct.omitted_field;
+            }
+
+            const AlignSortContext = struct {
+                struct_obj: *Module.Struct,
+                sema: *Sema,
+
+                fn lessThan(ctx: @This(), a: u32, b: u32) bool {
+                    if (a == Module.Struct.omitted_field) return false;
+                    if (b == Module.Struct.omitted_field) return true;
+                    const target = ctx.sema.mod.getTarget();
+                    return ctx.struct_obj.fields.values()[a].ty.abiAlignment(target) >
+                        ctx.struct_obj.fields.values()[b].ty.abiAlignment(target);
+                }
+            };
+            std.sort.sort(u32, optimized_order, AlignSortContext{
+                .struct_obj = struct_obj,
+                .sema = sema,
+            }, AlignSortContext.lessThan);
+            struct_obj.optimized_order = optimized_order.ptr;
+        }
     }
     // otherwise it's a tuple; no need to resolve anything
 }

src/codegen/llvm.zig

@@ -2083,15 +2083,15 @@ pub const Object = struct {
                 comptime assert(struct_layout_version == 2);
                 var offset: u64 = 0;
 
-                for (fields.values()) |field, i| {
-                    if (field.is_comptime or !field.ty.hasRuntimeBits()) continue;
-
+                var it = ty.castTag(.@"struct").?.data.runtimeFieldIterator();
+                while (it.next()) |field_and_index| {
+                    const field = field_and_index.field;
                     const field_size = field.ty.abiSize(target);
                     const field_align = field.alignment(target, layout);
                     const field_offset = std.mem.alignForwardGeneric(u64, offset, field_align);
                     offset = field_offset + field_size;
 
-                    const field_name = try gpa.dupeZ(u8, fields.keys()[i]);
+                    const field_name = try gpa.dupeZ(u8, fields.keys()[field_and_index.index]);
                     defer gpa.free(field_name);
 
                     try di_fields.append(gpa, dib.createMemberType(
@@ -2985,9 +2985,9 @@ pub const DeclGen = struct {
                 var big_align: u32 = 1;
                 var any_underaligned_fields = false;
 
-                for (struct_obj.fields.values()) |field| {
-                    if (field.is_comptime or !field.ty.hasRuntimeBits()) continue;
-
+                var it = struct_obj.runtimeFieldIterator();
+                while (it.next()) |field_and_index| {
+                    const field = field_and_index.field;
                     const field_align = field.alignment(target, struct_obj.layout);
                     const field_ty_align = field.ty.abiAlignment(target);
                     any_underaligned_fields = any_underaligned_fields or
@@ -3714,9 +3714,9 @@ pub const DeclGen = struct {
                 var big_align: u32 = 0;
                 var need_unnamed = false;
 
-                for (struct_obj.fields.values()) |field, i| {
-                    if (field.is_comptime or !field.ty.hasRuntimeBits()) continue;
-
+                var it = struct_obj.runtimeFieldIterator();
+                while (it.next()) |field_and_index| {
+                    const field = field_and_index.field;
                     const field_align = field.alignment(target, struct_obj.layout);
                     big_align = @max(big_align, field_align);
                     const prev_offset = offset;
@@ -3732,7 +3732,7 @@ pub const DeclGen = struct {
 
                     const field_llvm_val = try dg.lowerValue(.{
                         .ty = field.ty,
-                        .val = field_vals[i],
+                        .val = field_vals[field_and_index.index],
                     });
 
                     need_unnamed = need_unnamed or dg.isUnnamedType(field.ty, field_llvm_val);
@@ -10354,9 +10354,9 @@ fn llvmFieldIndex(
     assert(layout != .Packed);
 
     var llvm_field_index: c_uint = 0;
-    for (ty.structFields().values()) |field, i| {
-        if (field.is_comptime or !field.ty.hasRuntimeBits()) continue;
-
+    var it = ty.castTag(.@"struct").?.data.runtimeFieldIterator();
+    while (it.next()) |field_and_index| {
+        const field = field_and_index.field;
         const field_align = field.alignment(target, layout);
         big_align = @max(big_align, field_align);
         const prev_offset = offset;
@@ -10367,7 +10367,7 @@ fn llvmFieldIndex(
             llvm_field_index += 1;
         }
 
-        if (field_index <= i) {
+        if (field_index == field_and_index.index) {
             ptr_pl_buf.* = .{
                 .data = .{
                     .pointee_type = field.ty,

src/type.zig

@@ -5741,10 +5741,14 @@ pub const Type = extern union {
         target: Target,
 
         pub fn next(it: *StructOffsetIterator) ?FieldOffset {
-            const i = it.field;
+            var i = it.field;
             if (it.struct_obj.fields.count() <= i)
                 return null;
 
+            if (it.struct_obj.optimized_order) |some| {
+                i = some[i];
+                if (i == Module.Struct.omitted_field) return null;
+            }
             const field = it.struct_obj.fields.values()[i];
             it.field += 1;
 

test/behavior/struct.zig

@@ -1555,3 +1555,21 @@ test "optional generic function label struct field" {
     };
     try expect((Options{}).isFoo.?(u8) == 123);
 }
+
+test "struct fields get automatically reordered" {
+    if (builtin.zig_backend != .stage2_llvm) return error.SkipZigTest; // TODO
+
+    const S1 = struct {
+        a: u32,
+        b: u32,
+        c: bool,
+        d: bool,
+    };
+    const S2 = struct {
+        a: u32,
+        b: bool,
+        c: u32,
+        d: bool,
+    };
+    try expect(@sizeOf(S1) == @sizeOf(S2));
+}