Merge pull request ziglang#18155 from Luukdegram/wasm-gc

wasm-linker: implement garbage-collection and performance improvements

src/link/Wasm/Atom.zig

@@ -23,6 +23,10 @@ alignment: Wasm.Alignment,
 /// Offset into the section where the atom lives, this already accounts
 /// for alignment.
 offset: u32,
+/// The original offset within the object file. This value is substracted from
+/// relocation offsets to determine where in the `data` to rewrite the value
+original_offset: u32,
+
 /// Represents the index of the file this atom was generated from.
 /// This is 'null' when the atom was generated by a Decl from Zig code.
 file: ?u16,
@@ -50,11 +54,11 @@ pub const empty: Atom = .{
     .prev = null,
     .size = 0,
     .sym_index = 0,
+    .original_offset = 0,
 };
 
 /// Frees all resources owned by this `Atom`.
-pub fn deinit(atom: *Atom, wasm: *Wasm) void {
-    const gpa = wasm.base.allocator;
+pub fn deinit(atom: *Atom, gpa: std.mem.Allocator) void {
     atom.relocs.deinit(gpa);
     atom.code.deinit(gpa);
     atom.locals.deinit(gpa);
@@ -114,10 +118,10 @@ pub fn resolveRelocs(atom: *Atom, wasm_bin: *const Wasm) void {
             .R_WASM_GLOBAL_INDEX_I32,
             .R_WASM_MEMORY_ADDR_I32,
             .R_WASM_SECTION_OFFSET_I32,
-            => std.mem.writeInt(u32, atom.code.items[reloc.offset..][0..4], @as(u32, @intCast(value)), .little),
+            => std.mem.writeInt(u32, atom.code.items[reloc.offset - atom.original_offset ..][0..4], @as(u32, @intCast(value)), .little),
             .R_WASM_TABLE_INDEX_I64,
             .R_WASM_MEMORY_ADDR_I64,
-            => std.mem.writeInt(u64, atom.code.items[reloc.offset..][0..8], value, .little),
+            => std.mem.writeInt(u64, atom.code.items[reloc.offset - atom.original_offset ..][0..8], value, .little),
             .R_WASM_GLOBAL_INDEX_LEB,
             .R_WASM_EVENT_INDEX_LEB,
             .R_WASM_FUNCTION_INDEX_LEB,
@@ -127,12 +131,12 @@ pub fn resolveRelocs(atom: *Atom, wasm_bin: *const Wasm) void {
             .R_WASM_TABLE_NUMBER_LEB,
             .R_WASM_TYPE_INDEX_LEB,
             .R_WASM_MEMORY_ADDR_TLS_SLEB,
-            => leb.writeUnsignedFixed(5, atom.code.items[reloc.offset..][0..5], @as(u32, @intCast(value))),
+            => leb.writeUnsignedFixed(5, atom.code.items[reloc.offset - atom.original_offset ..][0..5], @as(u32, @intCast(value))),
             .R_WASM_MEMORY_ADDR_LEB64,
             .R_WASM_MEMORY_ADDR_SLEB64,
             .R_WASM_TABLE_INDEX_SLEB64,
             .R_WASM_MEMORY_ADDR_TLS_SLEB64,
-            => leb.writeUnsignedFixed(10, atom.code.items[reloc.offset..][0..10], value),
+            => leb.writeUnsignedFixed(10, atom.code.items[reloc.offset - atom.original_offset ..][0..10], value),
         }
     }
 }
@@ -150,7 +154,7 @@ fn relocationValue(atom: Atom, relocation: types.Relocation, wasm_bin: *const Wa
         .R_WASM_TABLE_INDEX_I64,
         .R_WASM_TABLE_INDEX_SLEB,
         .R_WASM_TABLE_INDEX_SLEB64,
-        => return wasm_bin.function_table.get(target_loc) orelse 0,
+        => return wasm_bin.function_table.get(.{ .file = atom.file, .index = relocation.index }) orelse 0,
         .R_WASM_TYPE_INDEX_LEB => {
             const file_index = atom.file orelse {
                 return relocation.index;

src/link/Wasm/Symbol.zig

@@ -79,6 +79,9 @@ pub const Flag = enum(u32) {
     WASM_SYM_NO_STRIP = 0x80,
     /// Indicates a symbol is TLS
     WASM_SYM_TLS = 0x100,
+    /// Zig specific flag. Uses the most significant bit of the flag to annotate whether a symbol is
+    /// alive or not. Dead symbols are allowed to be garbage collected.
+    alive = 0x80000000,
 };
 
 /// Verifies if the given symbol should be imported from the
@@ -92,6 +95,23 @@ pub fn requiresImport(symbol: Symbol) bool {
     return true;
 }
 
+/// Marks a symbol as 'alive', ensuring the garbage collector will not collect the trash.
+pub fn mark(symbol: *Symbol) void {
+    symbol.flags |= @intFromEnum(Flag.alive);
+}
+
+pub fn unmark(symbol: *Symbol) void {
+    symbol.flags &= ~@intFromEnum(Flag.alive);
+}
+
+pub fn isAlive(symbol: Symbol) bool {
+    return symbol.flags & @intFromEnum(Flag.alive) != 0;
+}
+
+pub fn isDead(symbol: Symbol) bool {
+    return symbol.flags & @intFromEnum(Flag.alive) == 0;
+}
+
 pub fn isTLS(symbol: Symbol) bool {
     return symbol.flags & @intFromEnum(Flag.WASM_SYM_TLS) != 0;
 }

test/link/wasm/bss/build.zig

@@ -26,6 +26,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize_mode: std.builtin.Opt
         lib.strip = false;
         // to make sure the bss segment is emitted, we must import memory
         lib.import_memory = true;
+        lib.link_gc_sections = false;
 
         const check_lib = lib.checkObject();
 
@@ -73,6 +74,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize_mode: std.builtin.Opt
         lib.strip = false;
         // to make sure the bss segment is emitted, we must import memory
         lib.import_memory = true;
+        lib.link_gc_sections = false;
 
         const check_lib = lib.checkObject();
         check_lib.checkStart();

test/link/wasm/function-table/build.zig

@@ -23,6 +23,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize: std.builtin.Optimize
     import_table.use_llvm = false;
     import_table.use_lld = false;
     import_table.import_table = true;
+    import_table.link_gc_sections = false;
 
     const export_table = b.addExecutable(.{
         .name = "export_table",
@@ -34,6 +35,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize: std.builtin.Optimize
     export_table.use_llvm = false;
     export_table.use_lld = false;
     export_table.export_table = true;
+    export_table.link_gc_sections = false;
 
     const regular_table = b.addExecutable(.{
         .name = "regular_table",
@@ -44,6 +46,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize: std.builtin.Optimize
     regular_table.entry = .disabled;
     regular_table.use_llvm = false;
     regular_table.use_lld = false;
+    regular_table.link_gc_sections = false; // Ensure function table is not empty
 
     const check_import = import_table.checkObject();
     const check_export = export_table.checkObject();

test/link/wasm/segments/build.zig

@@ -23,6 +23,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize: std.builtin.Optimize
     lib.use_llvm = false;
     lib.use_lld = false;
     lib.strip = false;
+    lib.link_gc_sections = false; // so data is not garbage collected and we can verify data section
     b.installArtifact(lib);
 
     const check_lib = lib.checkObject();

test/link/wasm/stack_pointer/build.zig

@@ -24,6 +24,7 @@ fn add(b: *std.Build, test_step: *std.Build.Step, optimize: std.builtin.Optimize
     lib.use_lld = false;
     lib.strip = false;
     lib.stack_size = std.wasm.page_size * 2; // set an explicit stack size
+    lib.link_gc_sections = false;
     b.installArtifact(lib);
 
     const check_lib = lib.checkObject();