authorgravatar for luuk@degram.devLuuk de Gram <luuk@degram.dev> 2023-11-26 18:55:06+01:00
committergravatar for luuk@degram.devLuuk de Gram <luuk@degram.dev> 2023-11-28 16:40:26+01:00
log596d1cd5a8a2daad25df9d39ad384df1b67eb39e
tree32c2b130127548b829e768ea227bba30eec26fe0
parent4be3cd2754f4af4f18b969dd60d954c53a281a83
signaturelock-open Commit is signed but in an unrecognized format.

wasm-linker: support `--no-gc-sections`

By default we garbage-collect sections for Wasm to reduce size, as well as finish linking quicker (as we have fewer things to do). However, when the user specifies `--no-gc-sections` we ensure all resolved symbols get marked and therefore do not get garbage collected. This is supported in both incremental-mode and traditional linking.

2 files changed, 59 insertions(+), 36 deletions(-)

src/link/Wasm.zig+58-35
......@@ -2134,10 +2134,14 @@ const Kind = union(enum) {
21342134fn parseAtom(wasm: *Wasm, atom_index: Atom.Index, kind: Kind) !void {
21352135 const atom = wasm.getAtomPtr(atom_index);
21362136 const symbol = (SymbolLoc{ .file = null, .index = atom.sym_index }).getSymbol(wasm);
2137 if (symbol.isDead()) {
2137 const do_garbage_collect = wasm.base.options.gc_sections orelse
2138 (wasm.base.options.output_mode != .Obj);
2139
2140 if (symbol.isDead() and do_garbage_collect) {
21382141 // Prevent unreferenced symbols from being parsed.
21392142 return;
21402143 }
2144
21412145 const final_index: u32 = switch (kind) {
21422146 .function => result: {
21432147 const index: u32 = @intCast(wasm.functions.count() + wasm.imported_functions_count);
......@@ -2289,16 +2293,23 @@ fn allocateAtoms(wasm: *Wasm) !void {
22892293 if (sym.isDead()) {
22902294 // Dead symbols must be unlinked from the linked-list to prevent them
22912295 // from being emit into the binary.
2292 if (atom.prev) |prev_index| {
2293 const prev = wasm.getAtomPtr(prev_index);
2294 prev.next = atom.next;
2296 if (atom.next) |next_index| {
2297 const next = wasm.getAtomPtr(next_index);
2298 next.prev = atom.prev;
2299 } else if (entry.value_ptr.* == atom_index) {
2300 // When the atom is dead and is also the first atom retrieved from wasm.atoms(index) we update
2301 // the entry to point it to the previous atom to ensure we do not start with a dead symbol that
2302 // was removed and therefore do not emit any code at all.
2303 if (atom.prev) |prev| {
2304 entry.value_ptr.* = prev;
2305 }
22952306 }
2296 atom_index = atom.next orelse {
2297 atom.prev = null;
2307 atom_index = atom.prev orelse {
2308 atom.next = null;
22982309 break;
22992310 };
2300 const next = wasm.getAtomPtr(atom_index);
2301 next.prev = atom.prev;
2311 const prev = wasm.getAtomPtr(atom_index);
2312 prev.next = atom.next;
23022313 atom.prev = null;
23032314 atom.next = null;
23042315 continue;
......@@ -2787,6 +2798,7 @@ fn mergeSections(wasm: *Wasm) !void {
27872798 // We found an alias to the same function, discard this symbol in favor of
27882799 // the original symbol and point the discard function to it. This ensures
27892800 // we only emit a single function, instead of duplicates.
2801 symbol.unmark();
27902802 try wasm.discarded.putNoClobber(
27912803 wasm.base.allocator,
27922804 sym_loc,
......@@ -2815,7 +2827,6 @@ fn mergeSections(wasm: *Wasm) !void {
28152827 // For any removed duplicates, remove them from the resolved symbols list
28162828 for (removed_duplicates.items) |sym_loc| {
28172829 assert(wasm.resolved_symbols.swapRemove(sym_loc));
2818 sym_loc.getSymbol(wasm).unmark();
28192830 }
28202831
28212832 log.debug("Merged ({d}) functions", .{wasm.functions.count()});
......@@ -3741,8 +3752,8 @@ fn writeToFile(
37413752 binary_bytes.items,
37423753 header_offset,
37433754 .type,
3744 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3745 @as(u32, @intCast(wasm.func_types.items.len)),
3755 @intCast(binary_bytes.items.len - header_offset - header_size),
3756 @intCast(wasm.func_types.items.len),
37463757 );
37473758 section_count += 1;
37483759 }
......@@ -3774,8 +3785,8 @@ fn writeToFile(
37743785 binary_bytes.items,
37753786 header_offset,
37763787 .import,
3777 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3778 @as(u32, @intCast(wasm.imports.count() + @intFromBool(import_memory))),
3788 @intCast(binary_bytes.items.len - header_offset - header_size),
3789 @intCast(wasm.imports.count() + @intFromBool(import_memory)),
37793790 );
37803791 section_count += 1;
37813792 }
......@@ -3791,8 +3802,8 @@ fn writeToFile(
37913802 binary_bytes.items,
37923803 header_offset,
37933804 .function,
3794 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3795 @as(u32, @intCast(wasm.functions.count())),
3805 @intCast(binary_bytes.items.len - header_offset - header_size),
3806 @intCast(wasm.functions.count()),
37963807 );
37973808 section_count += 1;
37983809 }
......@@ -3810,8 +3821,8 @@ fn writeToFile(
38103821 binary_bytes.items,
38113822 header_offset,
38123823 .table,
3813 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3814 @as(u32, @intCast(wasm.tables.items.len)),
3824 @intCast(binary_bytes.items.len - header_offset - header_size),
3825 @intCast(wasm.tables.items.len),
38153826 );
38163827 section_count += 1;
38173828 }
......@@ -3825,8 +3836,8 @@ fn writeToFile(
38253836 binary_bytes.items,
38263837 header_offset,
38273838 .memory,
3828 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3829 @as(u32, 1), // wasm currently only supports 1 linear memory segment
3839 @intCast(binary_bytes.items.len - header_offset - header_size),
3840 1, // wasm currently only supports 1 linear memory segment
38303841 );
38313842 section_count += 1;
38323843 }
......@@ -3845,8 +3856,8 @@ fn writeToFile(
38453856 binary_bytes.items,
38463857 header_offset,
38473858 .global,
3848 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3849 @as(u32, @intCast(wasm.wasm_globals.items.len)),
3859 @intCast(binary_bytes.items.len - header_offset - header_size),
3860 @intCast(wasm.wasm_globals.items.len),
38503861 );
38513862 section_count += 1;
38523863 }
......@@ -3874,8 +3885,8 @@ fn writeToFile(
38743885 binary_bytes.items,
38753886 header_offset,
38763887 .@"export",
3877 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3878 @as(u32, @intCast(wasm.exports.items.len)) + @intFromBool(export_memory),
3888 @intCast(binary_bytes.items.len - header_offset - header_size),
3889 @intCast(wasm.exports.items.len + @intFromBool(export_memory)),
38793890 );
38803891 section_count += 1;
38813892 }
......@@ -3918,8 +3929,8 @@ fn writeToFile(
39183929 binary_bytes.items,
39193930 header_offset,
39203931 .element,
3921 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3922 @as(u32, 1),
3932 @intCast(binary_bytes.items.len - header_offset - header_size),
3933 1,
39233934 );
39243935 section_count += 1;
39253936 }
......@@ -3932,8 +3943,8 @@ fn writeToFile(
39323943 binary_bytes.items,
39333944 header_offset,
39343945 .data_count,
3935 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
3936 @as(u32, @intCast(data_segments_count)),
3946 @intCast(binary_bytes.items.len - header_offset - header_size),
3947 @intCast(data_segments_count),
39373948 );
39383949 }
39393950
......@@ -3978,7 +3989,7 @@ fn writeToFile(
39783989 header_offset,
39793990 .code,
39803991 code_section_size,
3981 @as(u32, @intCast(wasm.functions.count())),
3992 @intCast(wasm.functions.count()),
39823993 );
39833994 code_section_index = section_count;
39843995 section_count += 1;
......@@ -4049,8 +4060,8 @@ fn writeToFile(
40494060 binary_bytes.items,
40504061 header_offset,
40514062 .data,
4052 @as(u32, @intCast(binary_bytes.items.len - header_offset - header_size)),
4053 @as(u32, @intCast(segment_count)),
4063 @intCast(binary_bytes.items.len - header_offset - header_size),
4064 @intCast(segment_count),
40544065 );
40554066 data_section_index = section_count;
40564067 section_count += 1;
......@@ -4597,6 +4608,14 @@ fn linkWithLLD(wasm: *Wasm, comp: *Compilation, prog_node: *std.Progress.Node) !
45974608 try argv.append("--export-table");
45984609 }
45994610
4611 if (wasm.base.options.gc_sections) |gc| {
4612 // For wasm-ld we only need to specify '--no-gc-sections' when the user explicitly
4613 // specified it as garbage collection is enabled by default.
4614 if (!gc) {
4615 try argv.append("--no-gc-sections");
4616 }
4617 }
4618
46004619 if (wasm.base.options.strip) {
46014620 try argv.append("-s");
46024621 }
......@@ -4882,7 +4901,7 @@ fn emitLinkSection(wasm: *Wasm, binary_bytes: *std.ArrayList(u8), symbol_table:
48824901 try wasm.emitSymbolTable(binary_bytes, symbol_table);
48834902 try wasm.emitSegmentInfo(binary_bytes);
48844903
4885 const size = @as(u32, @intCast(binary_bytes.items.len - offset - 6));
4904 const size: u32 = @intCast(binary_bytes.items.len - offset - 6);
48864905 try writeCustomSectionHeader(binary_bytes.items, offset, size);
48874906}
48884907
......@@ -4930,7 +4949,7 @@ fn emitSymbolTable(wasm: *Wasm, binary_bytes: *std.ArrayList(u8), symbol_table:
49304949 }
49314950
49324951 var buf: [10]u8 = undefined;
4933 leb.writeUnsignedFixed(5, buf[0..5], @as(u32, @intCast(binary_bytes.items.len - table_offset + 5)));
4952 leb.writeUnsignedFixed(5, buf[0..5], @intCast(binary_bytes.items.len - table_offset + 5));
49344953 leb.writeUnsignedFixed(5, buf[5..], symbol_count);
49354954 try binary_bytes.insertSlice(table_offset, &buf);
49364955}
......@@ -5013,7 +5032,7 @@ fn emitCodeRelocations(
50135032 var buf: [5]u8 = undefined;
50145033 leb.writeUnsignedFixed(5, &buf, count);
50155034 try binary_bytes.insertSlice(reloc_start, &buf);
5016 const size = @as(u32, @intCast(binary_bytes.items.len - header_offset - 6));
5035 const size: u32 = @intCast(binary_bytes.items.len - header_offset - 6);
50175036 try writeCustomSectionHeader(binary_bytes.items, header_offset, size);
50185037}
50195038
......@@ -5123,10 +5142,14 @@ pub fn storeDeclType(wasm: *Wasm, decl_index: InternPool.DeclIndex, func_type: s
51235142fn markReferences(wasm: *Wasm) !void {
51245143 const tracy = trace(@src());
51255144 defer tracy.end();
5145 const do_garbage_collect = wasm.base.options.gc_sections orelse
5146 (wasm.base.options.output_mode != .Obj);
5147
51265148 for (wasm.resolved_symbols.keys()) |sym_loc| {
51275149 const sym = sym_loc.getSymbol(wasm);
5128 if (sym.isExported(wasm.base.options.rdynamic) or sym.isNoStrip()) {
5150 if (sym.isExported(wasm.base.options.rdynamic) or sym.isNoStrip() or !do_garbage_collect) {
51295151 try wasm.mark(sym_loc);
5152 continue;
51305153 }
51315154
51325155 // Debug sections may require to be parsed and marked when it contains
......@@ -5139,7 +5162,7 @@ fn markReferences(wasm: *Wasm) !void {
51395162 for (atom.relocs.items) |reloc| {
51405163 const target_loc: SymbolLoc = .{ .index = reloc.index, .file = atom.file };
51415164 const target_sym = target_loc.getSymbol(wasm);
5142 if (target_sym.isAlive()) {
5165 if (target_sym.isAlive() or !do_garbage_collect) {
51435166 sym.mark();
51445167 continue; // Skip all other relocations as this debug atom is already marked now
51455168 }
src/link/Wasm/Object.zig+1-1
......@@ -920,7 +920,7 @@ pub fn parseSymbolIntoAtom(object: *Object, object_index: u16, symbol_index: u32
920920 const start = searchRelocStart(relocations, relocatable_data.offset);
921921 const len = searchRelocEnd(relocations[start..], relocatable_data.offset + atom.size);
922922 atom.relocs = std.ArrayListUnmanaged(types.Relocation).fromOwnedSlice(relocations[start..][0..len]);
923 for (atom.relocs.items) |*reloc| {
923 for (atom.relocs.items) |reloc| {
924924 switch (reloc.relocation_type) {
925925 .R_WASM_TABLE_INDEX_I32,
926926 .R_WASM_TABLE_INDEX_I64,