(wip) update wasm linker to new Writer API

This commit is contained in:
Andrew Kelley 2025-08-10 21:39:17 -07:00
parent 3280fc98f3
commit 168da23d8f
6 changed files with 873 additions and 1046 deletions

View File

@ -4,6 +4,7 @@ const std = @import("std");
const assert = std.debug.assert;
const Allocator = std.mem.Allocator;
const leb = std.leb;
const Writer = std.Io.Writer;
const Wasm = link.File.Wasm;
const Mir = @import("Mir.zig");
@ -14,16 +15,16 @@ const codegen = @import("../../codegen.zig");
mir: Mir,
wasm: *Wasm,
/// The binary representation that will be emitted by this module.
code: *std.ArrayListUnmanaged(u8),
/// The binary representation of this module is written here.
writer: *Writer,
pub const Error = error{
OutOfMemory,
WriteFailed,
};
pub fn lowerToCode(emit: *Emit) Error!void {
const mir = &emit.mir;
const code = emit.code;
const writer = emit.writer;
const wasm = emit.wasm;
const comp = wasm.base.comp;
const gpa = comp.gpa;
@ -41,18 +42,19 @@ pub fn lowerToCode(emit: *Emit) Error!void {
},
.block, .loop => {
const block_type = datas[inst].block_type;
try code.ensureUnusedCapacity(gpa, 2);
code.appendAssumeCapacity(@intFromEnum(tags[inst]));
code.appendAssumeCapacity(@intFromEnum(block_type));
try writer.writeAll(&.{
@intFromEnum(tags[inst]),
@intFromEnum(block_type),
});
inst += 1;
continue :loop tags[inst];
},
.uav_ref => {
if (is_obj) {
try uavRefObj(wasm, code, datas[inst].ip_index, 0, is_wasm32);
try uavRefObj(wasm, writer, datas[inst].ip_index, 0, is_wasm32);
} else {
try uavRefExe(wasm, code, datas[inst].ip_index, 0, is_wasm32);
try uavRefExe(wasm, writer, datas[inst].ip_index, 0, is_wasm32);
}
inst += 1;
continue :loop tags[inst];
@ -60,20 +62,20 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.uav_ref_off => {
const extra = mir.extraData(Mir.UavRefOff, datas[inst].payload).data;
if (is_obj) {
try uavRefObj(wasm, code, extra.value, extra.offset, is_wasm32);
try uavRefObj(wasm, writer, extra.value, extra.offset, is_wasm32);
} else {
try uavRefExe(wasm, code, extra.value, extra.offset, is_wasm32);
try uavRefExe(wasm, writer, extra.value, extra.offset, is_wasm32);
}
inst += 1;
continue :loop tags[inst];
},
.nav_ref => {
try navRefOff(wasm, code, .{ .nav_index = datas[inst].nav_index, .offset = 0 }, is_wasm32);
try navRefOff(wasm, writer, .{ .nav_index = datas[inst].nav_index, .offset = 0 }, is_wasm32);
inst += 1;
continue :loop tags[inst];
},
.nav_ref_off => {
try navRefOff(wasm, code, mir.extraData(Mir.NavRefOff, datas[inst].payload).data, is_wasm32);
try navRefOff(wasm, writer, mir.extraData(Mir.NavRefOff, datas[inst].payload).data, is_wasm32);
inst += 1;
continue :loop tags[inst];
},
@ -81,11 +83,11 @@ pub fn lowerToCode(emit: *Emit) Error!void {
const indirect_func_idx: Wasm.ZcuIndirectFunctionSetIndex = @enumFromInt(
wasm.zcu_indirect_function_set.getIndex(datas[inst].nav_index).?,
);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_const));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_const));
if (is_obj) {
@panic("TODO");
} else {
leb.writeUleb128(code.fixedWriter(), 1 + @intFromEnum(indirect_func_idx)) catch unreachable;
try writer.writeLeb128(1 + @intFromEnum(indirect_func_idx));
}
inst += 1;
continue :loop tags[inst];
@ -95,52 +97,48 @@ pub fn lowerToCode(emit: *Emit) Error!void {
continue :loop tags[inst];
},
.errors_len => {
try code.ensureUnusedCapacity(gpa, 6);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_const));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_const));
// MIR is lowered during flush, so there is indeed only one thread at this time.
const errors_len = 1 + comp.zcu.?.intern_pool.global_error_set.getNamesFromMainThread().len;
leb.writeIleb128(code.fixedWriter(), errors_len) catch unreachable;
const errors_len: u32 = @intCast(1 + comp.zcu.?.intern_pool.global_error_set.getNamesFromMainThread().len);
try writer.writeLeb128(@as(i32, @bitCast(errors_len)));
inst += 1;
continue :loop tags[inst];
},
.error_name_table_ref => {
wasm.error_name_table_ref_count += 1;
try code.ensureUnusedCapacity(gpa, 11);
const opcode: std.wasm.Opcode = if (is_wasm32) .i32_const else .i64_const;
code.appendAssumeCapacity(@intFromEnum(opcode));
try writer.writeByte(@intFromEnum(opcode));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.errorNameTableSymbolIndex() },
.tag = if (is_wasm32) .memory_addr_leb else .memory_addr_leb64,
.addend = 0,
});
code.appendNTimesAssumeCapacity(0, if (is_wasm32) 5 else 10);
try writer.splatByteAll(0, if (is_wasm32) 5 else 10);
inst += 1;
continue :loop tags[inst];
} else {
const addr: u32 = wasm.errorNameTableAddr();
leb.writeIleb128(code.fixedWriter(), addr) catch unreachable;
try writer.writeLeb128(@as(i32, @bitCast(addr)));
inst += 1;
continue :loop tags[inst];
}
},
.br_if, .br, .memory_grow, .memory_size => {
try code.ensureUnusedCapacity(gpa, 11);
code.appendAssumeCapacity(@intFromEnum(tags[inst]));
leb.writeUleb128(code.fixedWriter(), datas[inst].label) catch unreachable;
try writer.writeByte(@intFromEnum(tags[inst]));
try writer.writeLeb128(datas[inst].label);
inst += 1;
continue :loop tags[inst];
},
.local_get, .local_set, .local_tee => {
try code.ensureUnusedCapacity(gpa, 11);
code.appendAssumeCapacity(@intFromEnum(tags[inst]));
leb.writeUleb128(code.fixedWriter(), datas[inst].local) catch unreachable;
try writer.writeByte(@intFromEnum(tags[inst]));
try writer.writeLeb128(datas[inst].local);
inst += 1;
continue :loop tags[inst];
@ -150,29 +148,27 @@ pub fn lowerToCode(emit: *Emit) Error!void {
const extra_index = datas[inst].payload;
const extra = mir.extraData(Mir.JumpTable, extra_index);
const labels = mir.extra[extra.end..][0..extra.data.length];
try code.ensureUnusedCapacity(gpa, 11 + 10 * labels.len);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.br_table));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.br_table));
// -1 because default label is not part of length/depth.
leb.writeUleb128(code.fixedWriter(), extra.data.length - 1) catch unreachable;
for (labels) |label| leb.writeUleb128(code.fixedWriter(), label) catch unreachable;
try writer.writeLeb128(extra.data.length - 1);
for (labels) |label| try writer.writeLeb128(label);
inst += 1;
continue :loop tags[inst];
},
.call_nav => {
try code.ensureUnusedCapacity(gpa, 6);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.call));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.call));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.navSymbolIndex(datas[inst].nav_index) },
.tag = .function_index_leb,
.addend = 0,
});
code.appendNTimesAssumeCapacity(0, 5);
try writer.splatByteAll(0, 5);
} else {
appendOutputFunctionIndex(code, .fromIpNav(wasm, datas[inst].nav_index));
try appendOutputFunctionIndex(writer, .fromIpNav(wasm, datas[inst].nav_index));
}
inst += 1;
@ -180,7 +176,6 @@ pub fn lowerToCode(emit: *Emit) Error!void {
},
.call_indirect => {
try code.ensureUnusedCapacity(gpa, 11);
const fn_info = comp.zcu.?.typeToFunc(.fromInterned(datas[inst].ip_index)).?;
const func_ty_index = wasm.getExistingFunctionType(
fn_info.cc,
@ -188,38 +183,37 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.fromInterned(fn_info.return_type),
target,
).?;
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.call_indirect));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.call_indirect));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .type_index = func_ty_index },
.tag = .type_index_leb,
.addend = 0,
});
code.appendNTimesAssumeCapacity(0, 5);
try writer.splatByteAll(0, 5);
} else {
const index: Wasm.Flush.FuncTypeIndex = .fromTypeIndex(func_ty_index, &wasm.flush_buffer);
leb.writeUleb128(code.fixedWriter(), @intFromEnum(index)) catch unreachable;
try writer.writeLeb128(@intFromEnum(index));
}
leb.writeUleb128(code.fixedWriter(), @as(u32, 0)) catch unreachable; // table index
try writer.writeUleb128(0); // table index
inst += 1;
continue :loop tags[inst];
},
.call_tag_name => {
try code.ensureUnusedCapacity(gpa, 6);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.call));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.call));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.tagNameSymbolIndex(datas[inst].ip_index) },
.tag = .function_index_leb,
.addend = 0,
});
code.appendNTimesAssumeCapacity(0, 5);
try writer.splatByteAll(0, 5);
} else {
appendOutputFunctionIndex(code, .fromTagNameType(wasm, datas[inst].ip_index));
try appendOutputFunctionIndex(writer, .fromTagNameType(wasm, datas[inst].ip_index));
}
inst += 1;
@ -232,18 +226,17 @@ pub fn lowerToCode(emit: *Emit) Error!void {
// table initialized based on the `Mir.Intrinsic` enum.
const symbol_name = try wasm.internString(@tagName(datas[inst].intrinsic));
try code.ensureUnusedCapacity(gpa, 6);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.call));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.call));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.symbolNameIndex(symbol_name) },
.tag = .function_index_leb,
.addend = 0,
});
code.appendNTimesAssumeCapacity(0, 5);
try writer.splatByteAll(0, 5);
} else {
appendOutputFunctionIndex(code, .fromSymbolName(wasm, symbol_name));
try appendOutputFunctionIndex(writer, .fromSymbolName(wasm, symbol_name));
}
inst += 1;
@ -251,19 +244,17 @@ pub fn lowerToCode(emit: *Emit) Error!void {
},
.global_set_sp => {
try code.ensureUnusedCapacity(gpa, 6);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.global_set));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.global_set));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.stackPointerSymbolIndex() },
.tag = .global_index_leb,
.addend = 0,
});
code.appendNTimesAssumeCapacity(0, 5);
try writer.splatByteAll(0, 5);
} else {
const sp_global: Wasm.GlobalIndex = .stack_pointer;
std.leb.writeUleb128(code.fixedWriter(), @intFromEnum(sp_global)) catch unreachable;
try writer.writeLeb128(@intFromEnum(Wasm.GlobalIndex.stack_pointer));
}
inst += 1;
@ -271,36 +262,32 @@ pub fn lowerToCode(emit: *Emit) Error!void {
},
.f32_const => {
try code.ensureUnusedCapacity(gpa, 5);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.f32_const));
std.mem.writeInt(u32, code.addManyAsArrayAssumeCapacity(4), @bitCast(datas[inst].float32), .little);
try writer.writeByte(@intFromEnum(std.wasm.Opcode.f32_const));
try writer.writeInt(u32, @bitCast(datas[inst].float32), .little);
inst += 1;
continue :loop tags[inst];
},
.f64_const => {
try code.ensureUnusedCapacity(gpa, 9);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.f64_const));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.f64_const));
const float64 = mir.extraData(Mir.Float64, datas[inst].payload).data;
std.mem.writeInt(u64, code.addManyAsArrayAssumeCapacity(8), float64.toInt(), .little);
try writer.writeInt(u64, float64.toInt(), .little);
inst += 1;
continue :loop tags[inst];
},
.i32_const => {
try code.ensureUnusedCapacity(gpa, 6);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_const));
leb.writeIleb128(code.fixedWriter(), datas[inst].imm32) catch unreachable;
try writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_const));
try writer.writeLeb128(datas[inst].imm32);
inst += 1;
continue :loop tags[inst];
},
.i64_const => {
try code.ensureUnusedCapacity(gpa, 11);
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i64_const));
try writer.writeByte(@intFromEnum(std.wasm.Opcode.i64_const));
const int64: i64 = @bitCast(mir.extraData(Mir.Imm64, datas[inst].payload).data.toInt());
leb.writeIleb128(code.fixedWriter(), int64) catch unreachable;
try writer.writeLeb128(int64);
inst += 1;
continue :loop tags[inst];
@ -330,9 +317,8 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.i64_store16,
.i64_store32,
=> {
try code.ensureUnusedCapacity(gpa, 1 + 20);
code.appendAssumeCapacity(@intFromEnum(tags[inst]));
encodeMemArg(code, mir.extraData(Mir.MemArg, datas[inst].payload).data);
try writer.writeByte(@intFromEnum(tags[inst]));
try encodeMemArg(writer, mir.extraData(Mir.MemArg, datas[inst].payload).data);
inst += 1;
continue :loop tags[inst];
},
@ -466,43 +452,42 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.i64_clz,
.i64_ctz,
=> {
try code.append(gpa, @intFromEnum(tags[inst]));
try writer.writeByte(@intFromEnum(tags[inst]));
inst += 1;
continue :loop tags[inst];
},
.misc_prefix => {
try code.ensureUnusedCapacity(gpa, 6 + 6);
const extra_index = datas[inst].payload;
const opcode = mir.extra[extra_index];
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.misc_prefix));
leb.writeUleb128(code.fixedWriter(), opcode) catch unreachable;
switch (@as(std.wasm.MiscOpcode, @enumFromInt(opcode))) {
const opcode: std.wasm.MiscOpcode = @enumFromInt(mir.extra[extra_index]);
try writer.writeByte(@intFromEnum(std.wasm.Opcode.misc_prefix));
try writer.writeLeb128(@intFromEnum(opcode));
switch (opcode) {
// bulk-memory opcodes
.data_drop => {
const segment = mir.extra[extra_index + 1];
leb.writeUleb128(code.fixedWriter(), segment) catch unreachable;
try writer.writeLeb128(segment);
inst += 1;
continue :loop tags[inst];
},
.memory_init => {
const segment = mir.extra[extra_index + 1];
leb.writeUleb128(code.fixedWriter(), segment) catch unreachable;
leb.writeUleb128(code.fixedWriter(), @as(u32, 0)) catch unreachable; // memory index
try writer.writeLeb128(segment);
try writer.writeByte(0); // memory index
inst += 1;
continue :loop tags[inst];
},
.memory_fill => {
leb.writeUleb128(code.fixedWriter(), @as(u32, 0)) catch unreachable; // memory index
try writer.writeByte(0); // memory index
inst += 1;
continue :loop tags[inst];
},
.memory_copy => {
leb.writeUleb128(code.fixedWriter(), @as(u32, 0)) catch unreachable; // dst memory index
leb.writeUleb128(code.fixedWriter(), @as(u32, 0)) catch unreachable; // src memory index
try writer.writeByte(0); // dst memory index
try writer.writeByte(0); // src memory index
inst += 1;
continue :loop tags[inst];
@ -534,12 +519,11 @@ pub fn lowerToCode(emit: *Emit) Error!void {
comptime unreachable;
},
.simd_prefix => {
try code.ensureUnusedCapacity(gpa, 6 + 20);
const extra_index = datas[inst].payload;
const opcode = mir.extra[extra_index];
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.simd_prefix));
leb.writeUleb128(code.fixedWriter(), opcode) catch unreachable;
switch (@as(std.wasm.SimdOpcode, @enumFromInt(opcode))) {
const opcode: std.wasm.SimdOpcode = @enumFromInt(mir.extra[extra_index]);
try writer.writeByte(@intFromEnum(std.wasm.Opcode.simd_prefix));
try writer.writeLeb128(@intFromEnum(opcode));
switch (opcode) {
.v128_store,
.v128_load,
.v128_load8_splat,
@ -547,12 +531,12 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.v128_load32_splat,
.v128_load64_splat,
=> {
encodeMemArg(code, mir.extraData(Mir.MemArg, extra_index + 1).data);
try encodeMemArg(writer, mir.extraData(Mir.MemArg, extra_index + 1).data);
inst += 1;
continue :loop tags[inst];
},
.v128_const, .i8x16_shuffle => {
code.appendSliceAssumeCapacity(std.mem.asBytes(mir.extra[extra_index + 1 ..][0..4]));
try writer.writeAll(std.mem.asBytes(mir.extra[extra_index + 1 ..][0..4]));
inst += 1;
continue :loop tags[inst];
},
@ -571,7 +555,7 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.f64x2_extract_lane,
.f64x2_replace_lane,
=> {
code.appendAssumeCapacity(@intCast(mir.extra[extra_index + 1]));
try writer.writeByte(@intCast(mir.extra[extra_index + 1]));
inst += 1;
continue :loop tags[inst];
},
@ -819,13 +803,11 @@ pub fn lowerToCode(emit: *Emit) Error!void {
comptime unreachable;
},
.atomics_prefix => {
try code.ensureUnusedCapacity(gpa, 6 + 20);
const extra_index = datas[inst].payload;
const opcode = mir.extra[extra_index];
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.atomics_prefix));
leb.writeUleb128(code.fixedWriter(), opcode) catch unreachable;
switch (@as(std.wasm.AtomicsOpcode, @enumFromInt(opcode))) {
const opcode: std.wasm.AtomicsOpcode = @enumFromInt(mir.extra[extra_index]);
try writer.writeByte(@intFromEnum(std.wasm.Opcode.atomics_prefix));
try writer.writeLeb128(@intFromEnum(opcode));
switch (opcode) {
.i32_atomic_load,
.i64_atomic_load,
.i32_atomic_load8_u,
@ -892,15 +874,12 @@ pub fn lowerToCode(emit: *Emit) Error!void {
.i64_atomic_rmw32_cmpxchg_u,
=> {
const mem_arg = mir.extraData(Mir.MemArg, extra_index + 1).data;
encodeMemArg(code, mem_arg);
try encodeMemArg(writer, mem_arg);
inst += 1;
continue :loop tags[inst];
},
.atomic_fence => {
// Hard-codes memory index 0 since multi-memory proposal is
// not yet accepted nor implemented.
const memory_index: u32 = 0;
leb.writeUleb128(code.fixedWriter(), memory_index) catch unreachable;
try writer.writeByte(0); // memory index
inst += 1;
continue :loop tags[inst];
},
@ -915,44 +894,36 @@ pub fn lowerToCode(emit: *Emit) Error!void {
}
/// Asserts 20 unused capacity.
fn encodeMemArg(code: *std.ArrayListUnmanaged(u8), mem_arg: Mir.MemArg) void {
assert(code.unusedCapacitySlice().len >= 20);
// Wasm encodes alignment as power of 2, rather than natural alignment.
const encoded_alignment = @ctz(mem_arg.alignment);
leb.writeUleb128(code.fixedWriter(), encoded_alignment) catch unreachable;
leb.writeUleb128(code.fixedWriter(), mem_arg.offset) catch unreachable;
fn encodeMemArg(writer: *Writer, mem_arg: Mir.MemArg) Writer.Error!void {
try writer.writeLeb128(Wasm.Alignment.fromNonzeroByteUnits(mem_arg.alignment).toLog2Units());
try writer.writeLeb128(mem_arg.offset);
}
fn uavRefObj(wasm: *Wasm, code: *std.ArrayListUnmanaged(u8), value: InternPool.Index, offset: i32, is_wasm32: bool) !void {
fn uavRefObj(wasm: *Wasm, writer: *Writer, value: InternPool.Index, offset: i32, is_wasm32: bool) Writer.Error!void {
const comp = wasm.base.comp;
const gpa = comp.gpa;
const opcode: std.wasm.Opcode = if (is_wasm32) .i32_const else .i64_const;
try code.ensureUnusedCapacity(gpa, 11);
code.appendAssumeCapacity(@intFromEnum(opcode));
try writer.writeByte(@intFromEnum(opcode));
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.uavSymbolIndex(value) },
.tag = if (is_wasm32) .memory_addr_leb else .memory_addr_leb64,
.addend = offset,
});
code.appendNTimesAssumeCapacity(0, if (is_wasm32) 5 else 10);
try writer.splatByteAll(0, if (is_wasm32) 5 else 10);
}
fn uavRefExe(wasm: *Wasm, code: *std.ArrayListUnmanaged(u8), value: InternPool.Index, offset: i32, is_wasm32: bool) !void {
const comp = wasm.base.comp;
const gpa = comp.gpa;
fn uavRefExe(wasm: *Wasm, writer: *Writer, value: InternPool.Index, offset: i32, is_wasm32: bool) !void {
const opcode: std.wasm.Opcode = if (is_wasm32) .i32_const else .i64_const;
try code.ensureUnusedCapacity(gpa, 11);
code.appendAssumeCapacity(@intFromEnum(opcode));
try writer.writeByte(@intFromEnum(opcode));
const addr = wasm.uavAddr(value);
leb.writeUleb128(code.fixedWriter(), @as(u32, @intCast(@as(i64, addr) + offset))) catch unreachable;
try writer.writeLeb128(@as(u32, @intCast(@as(i64, addr) + offset)));
}
fn navRefOff(wasm: *Wasm, code: *std.ArrayListUnmanaged(u8), data: Mir.NavRefOff, is_wasm32: bool) !void {
fn navRefOff(wasm: *Wasm, writer: *Writer, data: Mir.NavRefOff, is_wasm32: bool) !void {
const comp = wasm.base.comp;
const zcu = comp.zcu.?;
const ip = &zcu.intern_pool;
@ -961,24 +932,22 @@ fn navRefOff(wasm: *Wasm, code: *std.ArrayListUnmanaged(u8), data: Mir.NavRefOff
const nav_ty = ip.getNav(data.nav_index).typeOf(ip);
assert(!ip.isFunctionType(nav_ty));
try code.ensureUnusedCapacity(gpa, 11);
const opcode: std.wasm.Opcode = if (is_wasm32) .i32_const else .i64_const;
code.appendAssumeCapacity(@intFromEnum(opcode));
try writer.writeByte(@intFromEnum(opcode));
if (is_obj) {
try wasm.out_relocs.append(gpa, .{
.offset = @intCast(code.items.len),
.offset = @intCast(writer.count),
.pointee = .{ .symbol_index = try wasm.navSymbolIndex(data.nav_index) },
.tag = if (is_wasm32) .memory_addr_leb else .memory_addr_leb64,
.addend = data.offset,
});
code.appendNTimesAssumeCapacity(0, if (is_wasm32) 5 else 10);
try writer.splatByteAll(0, if (is_wasm32) 5 else 10);
} else {
const addr = wasm.navAddr(data.nav_index);
leb.writeUleb128(code.fixedWriter(), @as(u32, @intCast(@as(i64, addr) + data.offset))) catch unreachable;
try writer.writeLeb128(@as(i32, @bitCast(@as(u32, @intCast(@as(i64, addr) + data.offset)))));
}
}
fn appendOutputFunctionIndex(code: *std.ArrayListUnmanaged(u8), i: Wasm.OutputFunctionIndex) void {
leb.writeUleb128(code.fixedWriter(), @intFromEnum(i)) catch unreachable;
fn appendOutputFunctionIndex(writer: *Writer, i: Wasm.OutputFunctionIndex) Writer.Error!void {
return writer.writeLeb128(@intFromEnum(i));
}

View File

@ -669,16 +669,14 @@ pub fn deinit(mir: *Mir, gpa: std.mem.Allocator) void {
mir.* = undefined;
}
pub fn lower(mir: *const Mir, wasm: *Wasm, code: *std.ArrayListUnmanaged(u8)) std.mem.Allocator.Error!void {
const gpa = wasm.base.comp.gpa;
pub fn lower(mir: *const Mir, wasm: *Wasm, writer: *std.Io.Writer) std.Io.Writer.Error!void {
// Write the locals in the prologue of the function body.
try code.ensureUnusedCapacity(gpa, 5 + mir.locals.len * 6 + 38);
_ = try writer.writableSliceGreedy(5 + mir.locals.len * 6 + 38);
std.leb.writeUleb128(code.fixedWriter(), @as(u32, @intCast(mir.locals.len))) catch unreachable;
writer.writeLeb128(@as(u32, @intCast(mir.locals.len))) catch unreachable;
for (mir.locals) |local| {
std.leb.writeUleb128(code.fixedWriter(), @as(u32, 1)) catch unreachable;
code.appendAssumeCapacity(@intFromEnum(local));
writer.writeLeb128(@as(u32, 1)) catch unreachable;
writer.writeByte(@intFromEnum(local)) catch unreachable;
}
// Stack management section of function prologue.
@ -686,37 +684,37 @@ pub fn lower(mir: *const Mir, wasm: *Wasm, code: *std.ArrayListUnmanaged(u8)) st
if (stack_alignment.toByteUnits()) |align_bytes| {
const sp_global: Wasm.GlobalIndex = .stack_pointer;
// load stack pointer
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.global_get));
std.leb.writeUleb128(code.fixedWriter(), @intFromEnum(sp_global)) catch unreachable;
writer.writeByte(@intFromEnum(std.wasm.Opcode.global_get)) catch unreachable;
writer.writeLeb128(@intFromEnum(sp_global)) catch unreachable;
// store stack pointer so we can restore it when we return from the function
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.local_tee));
leb.writeUleb128(code.fixedWriter(), mir.prologue.sp_local) catch unreachable;
writer.writeByte(@intFromEnum(std.wasm.Opcode.local_tee)) catch unreachable;
writer.writeLeb128(mir.prologue.sp_local) catch unreachable;
// get the total stack size
const aligned_stack: i32 = @intCast(stack_alignment.forward(mir.prologue.stack_size));
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_const));
leb.writeIleb128(code.fixedWriter(), aligned_stack) catch unreachable;
writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_const)) catch unreachable;
writer.writeLeb128(aligned_stack) catch unreachable;
// subtract it from the current stack pointer
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_sub));
writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_sub)) catch unreachable;
// Get negative stack alignment
const neg_stack_align = @as(i32, @intCast(align_bytes)) * -1;
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_const));
leb.writeIleb128(code.fixedWriter(), neg_stack_align) catch unreachable;
writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_const)) catch unreachable;
writer.writeLeb128(neg_stack_align) catch unreachable;
// Bitwise-and the value to get the new stack pointer to ensure the
// pointers are aligned with the abi alignment.
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.i32_and));
writer.writeByte(@intFromEnum(std.wasm.Opcode.i32_and)) catch unreachable;
// The bottom will be used to calculate all stack pointer offsets.
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.local_tee));
leb.writeUleb128(code.fixedWriter(), mir.prologue.bottom_stack_local) catch unreachable;
writer.writeByte(@intFromEnum(std.wasm.Opcode.local_tee)) catch unreachable;
writer.writeLeb128(mir.prologue.bottom_stack_local) catch unreachable;
// Store the current stack pointer value into the global stack pointer so other function calls will
// start from this value instead and not overwrite the current stack.
code.appendAssumeCapacity(@intFromEnum(std.wasm.Opcode.global_set));
std.leb.writeUleb128(code.fixedWriter(), @intFromEnum(sp_global)) catch unreachable;
writer.writeByte(@intFromEnum(std.wasm.Opcode.global_set)) catch unreachable;
writer.writeLeb128(@intFromEnum(sp_global)) catch unreachable;
}
var emit: Emit = .{
.mir = mir.*,
.wasm = wasm,
.code = code,
.writer = writer,
};
try emit.lowerToCode();
}

View File

@ -28,6 +28,7 @@ const fs = std.fs;
const leb = std.leb;
const log = std.log.scoped(.link);
const mem = std.mem;
const Writer = std.Io.Writer;
const Mir = @import("../arch/wasm/Mir.zig");
const CodeGen = @import("../arch/wasm/CodeGen.zig");
@ -2087,11 +2088,9 @@ pub const Expr = enum(u32) {
pub const end = @intFromEnum(std.wasm.Opcode.end);
pub fn slice(index: Expr, wasm: *const Wasm) [:end]const u8 {
const start_slice = wasm.string_bytes.items[@intFromEnum(index)..];
const end_pos = Object.exprEndPos(start_slice, 0) catch |err| switch (err) {
error.InvalidInitOpcode => unreachable,
};
return start_slice[0..end_pos :end];
var r: std.Io.Reader = .fixed(wasm.string_bytes.items[@intFromEnum(index)..]);
Object.skipInit(&r) catch unreachable;
return r.buffered()[0 .. r.seek - 1 :end];
}
};
@ -2126,7 +2125,7 @@ pub const FunctionType = extern struct {
wasm: *const Wasm,
ft: FunctionType,
pub fn format(self: Formatter, writer: *std.io.Writer) std.io.Writer.Error!void {
pub fn format(self: Formatter, writer: *Writer) Writer.Error!void {
const params = self.ft.params.slice(self.wasm);
const returns = self.ft.returns.slice(self.wasm);
@ -2905,7 +2904,7 @@ pub const Feature = packed struct(u8) {
@"=",
};
pub fn format(feature: Feature, writer: *std.io.Writer) std.io.Writer.Error!void {
pub fn format(feature: Feature, writer: *Writer) Writer.Error!void {
try writer.print("{s} {s}", .{ @tagName(feature.prefix), @tagName(feature.tag) });
}
@ -3037,16 +3036,16 @@ fn parseObject(wasm: *Wasm, obj: link.Input.Object) !void {
const stat = try obj.file.stat();
const size = std.math.cast(usize, stat.size) orelse return error.FileTooBig;
const file_contents = try gpa.alloc(u8, size);
defer gpa.free(file_contents);
var br: std.Io.Reader = .fixed(try gpa.alloc(u8, size));
defer gpa.free(br.buffered());
const n = try obj.file.preadAll(file_contents, 0);
if (n != file_contents.len) return error.UnexpectedEndOfFile;
const n = try obj.file.preadAll(br.buffered(), 0);
if (n != br.bufferedLen()) return error.UnexpectedEndOfFile;
var ss: Object.ScratchSpace = .{};
defer ss.deinit(gpa);
const object = try Object.parse(wasm, file_contents, obj.path, null, wasm.object_host_name, &ss, obj.must_link, gc_sections);
const object = try Object.parse(wasm, &br, obj.path, null, wasm.object_host_name, &ss, obj.must_link, gc_sections);
wasm.objects.appendAssumeCapacity(object);
}

View File

@ -167,9 +167,8 @@ pub fn parseObject(
};
const object_file_size = try header.parsedSize();
const contents = file_contents[object_offset + @sizeOf(Header) ..][0..object_file_size];
return Object.parse(wasm, contents, path, object_name, host_name, scratch_space, must_link, gc_sections);
var r: std.io.Reader = .fixed(file_contents[object_offset + @sizeOf(Header) ..][0..object_file_size]);
return Object.parse(wasm, &r, path, object_name, host_name, scratch_space, must_link, gc_sections);
}
const Archive = @This();

File diff suppressed because it is too large Load Diff

View File

@ -8,6 +8,7 @@ const Allocator = std.mem.Allocator;
const Path = std.Build.Cache.Path;
const log = std.log.scoped(.object);
const assert = std.debug.assert;
const Reader = std.Io.Reader;
/// Wasm spec version used for this `Object`
version: u32,
@ -252,25 +253,21 @@ pub const ScratchSpace = struct {
pub fn parse(
wasm: *Wasm,
bytes: []const u8,
br: *Reader,
path: Path,
archive_member_name: ?[]const u8,
host_name: Wasm.OptionalString,
ss: *ScratchSpace,
must_link: bool,
gc_sections: bool,
) anyerror!Object {
) !Object {
const comp = wasm.base.comp;
const gpa = comp.gpa;
const diags = &comp.link_diags;
var pos: usize = 0;
if (!std.mem.eql(u8, try br.takeArray(std.wasm.magic.len), &std.wasm.magic)) return error.BadObjectMagic;
if (!std.mem.eql(u8, bytes[0..std.wasm.magic.len], &std.wasm.magic)) return error.BadObjectMagic;
pos += std.wasm.magic.len;
const version = std.mem.readInt(u32, bytes[pos..][0..4], .little);
pos += 4;
const version = try br.takeInt(u32, .little);
const data_segment_start: u32 = @intCast(wasm.object_data_segments.items.len);
const custom_segment_start: u32 = @intCast(wasm.object_custom_segments.entries.len);
@ -298,200 +295,187 @@ pub fn parse(
var code_section_index: ?Wasm.ObjectSectionIndex = null;
var global_section_index: ?Wasm.ObjectSectionIndex = null;
var data_section_index: ?Wasm.ObjectSectionIndex = null;
while (pos < bytes.len) : (wasm.object_total_sections += 1) {
while (br.takeEnum(std.wasm.Section, .little)) |section_tag| : (wasm.object_total_sections += 1) {
const section_index: Wasm.ObjectSectionIndex = @enumFromInt(wasm.object_total_sections);
const section_tag: std.wasm.Section = @enumFromInt(bytes[pos]);
pos += 1;
const len, pos = readLeb(u32, bytes, pos);
const section_end = pos + len;
const len = try br.takeLeb128(u32);
const section_end = br.seek + len;
switch (section_tag) {
.custom => {
const section_name, pos = readBytes(bytes, pos);
const section_name = try br.take(try br.takeLeb128(u32));
if (std.mem.eql(u8, section_name, "linking")) {
saw_linking_section = true;
const section_version, pos = readLeb(u32, bytes, pos);
const section_version = try br.takeLeb128(u32);
log.debug("link meta data version: {d}", .{section_version});
if (section_version != 2) return error.UnsupportedVersion;
while (pos < section_end) {
const sub_type, pos = readLeb(u8, bytes, pos);
log.debug("found subsection: {s}", .{@tagName(@as(SubsectionType, @enumFromInt(sub_type)))});
const payload_len, pos = readLeb(u32, bytes, pos);
while (br.seek < section_end) {
const sub_type = try br.takeEnum(SubsectionType, .little);
log.debug("found subsection: {s}", .{@tagName(sub_type)});
const payload_len = try br.takeLeb128(u32);
if (payload_len == 0) break;
const count, pos = readLeb(u32, bytes, pos);
switch (@as(SubsectionType, @enumFromInt(sub_type))) {
.segment_info => {
for (try ss.segment_info.addManyAsSlice(gpa, count)) |*segment| {
const name, pos = readBytes(bytes, pos);
const alignment, pos = readLeb(u32, bytes, pos);
const flags_u32, pos = readLeb(u32, bytes, pos);
const flags: SegmentInfo.Flags = @bitCast(flags_u32);
const tls = flags.tls or
// Supports legacy object files that specified
// being TLS by the name instead of the TLS flag.
std.mem.startsWith(u8, name, ".tdata") or
std.mem.startsWith(u8, name, ".tbss");
has_tls = has_tls or tls;
segment.* = .{
.name = try wasm.internString(name),
.flags = .{
.strings = flags.strings,
.tls = tls,
.alignment = @enumFromInt(alignment),
.retain = flags.retain,
},
};
}
const count = try br.takeLeb128(u32);
switch (sub_type) {
.segment_info => for (try ss.segment_info.addManyAsSlice(gpa, count)) |*segment| {
const name = try br.take(try br.takeLeb128(u32));
const alignment: Alignment = .fromLog2Units(try br.takeLeb128(u32));
const flags: SegmentInfo.Flags = @bitCast(try br.takeLeb128(u32));
const tls = flags.tls or
// Supports legacy object files that specified
// being TLS by the name instead of the TLS flag.
std.mem.startsWith(u8, name, ".tdata") or
std.mem.startsWith(u8, name, ".tbss");
has_tls = has_tls or tls;
segment.* = .{
.name = try wasm.internString(name),
.flags = .{
.strings = flags.strings,
.tls = tls,
.alignment = alignment,
.retain = flags.retain,
},
};
},
.init_funcs => {
for (try wasm.object_init_funcs.addManyAsSlice(gpa, count)) |*func| {
const priority, pos = readLeb(u32, bytes, pos);
const symbol_index, pos = readLeb(u32, bytes, pos);
if (symbol_index > ss.symbol_table.items.len)
return diags.failParse(path, "init_funcs before symbol table", .{});
const sym = &ss.symbol_table.items[symbol_index];
if (sym.pointee != .function) {
return diags.failParse(path, "init_func symbol '{s}' not a function", .{
sym.name.slice(wasm).?,
});
} else if (sym.flags.undefined) {
return diags.failParse(path, "init_func symbol '{s}' is an import", .{
sym.name.slice(wasm).?,
});
}
func.* = .{
.priority = priority,
.function_index = sym.pointee.function,
};
.init_funcs => for (try wasm.object_init_funcs.addManyAsSlice(gpa, count)) |*func| {
const priority = try br.takeLeb128(u32);
const symbol_index = try br.takeLeb128(u32);
if (symbol_index > ss.symbol_table.items.len)
return diags.failParse(path, "init_funcs before symbol table", .{});
const sym = &ss.symbol_table.items[symbol_index];
if (sym.pointee != .function) {
return diags.failParse(path, "init_func symbol '{s}' not a function", .{
sym.name.slice(wasm).?,
});
} else if (sym.flags.undefined) {
return diags.failParse(path, "init_func symbol '{s}' is an import", .{
sym.name.slice(wasm).?,
});
}
func.* = .{
.priority = priority,
.function_index = sym.pointee.function,
};
},
.comdat_info => {
for (try wasm.object_comdats.addManyAsSlice(gpa, count)) |*comdat| {
const name, pos = readBytes(bytes, pos);
const flags, pos = readLeb(u32, bytes, pos);
if (flags != 0) return error.UnexpectedComdatFlags;
const symbol_count, pos = readLeb(u32, bytes, pos);
const start_off: u32 = @intCast(wasm.object_comdat_symbols.len);
try wasm.object_comdat_symbols.ensureUnusedCapacity(gpa, symbol_count);
for (0..symbol_count) |_| {
const kind, pos = readEnum(Wasm.Comdat.Symbol.Type, bytes, pos);
const index, pos = readLeb(u32, bytes, pos);
if (true) @panic("TODO rebase index depending on kind");
wasm.object_comdat_symbols.appendAssumeCapacity(.{
.kind = kind,
.index = index,
});
}
comdat.* = .{
.name = try wasm.internString(name),
.flags = flags,
.symbols = .{
.off = start_off,
.len = @intCast(wasm.object_comdat_symbols.len - start_off),
},
};
.comdat_info => for (try wasm.object_comdats.addManyAsSlice(gpa, count)) |*comdat| {
const name = try br.take(try br.takeLeb128(u32));
const flags = try br.takeLeb128(u32);
if (flags != 0) return error.UnexpectedComdatFlags;
const symbol_count = try br.takeLeb128(u32);
const start_off: u32 = @intCast(wasm.object_comdat_symbols.len);
try wasm.object_comdat_symbols.ensureUnusedCapacity(gpa, symbol_count);
for (0..symbol_count) |_| {
const kind = try br.takeEnum(Wasm.Comdat.Symbol.Type, .little);
const index = try br.takeLeb128(u32);
if (true) @panic("TODO rebase index depending on kind");
wasm.object_comdat_symbols.appendAssumeCapacity(.{
.kind = kind,
.index = index,
});
}
comdat.* = .{
.name = try wasm.internString(name),
.flags = flags,
.symbols = .{
.off = start_off,
.len = @intCast(wasm.object_comdat_symbols.len - start_off),
},
};
},
.symbol_table => {
for (try ss.symbol_table.addManyAsSlice(gpa, count)) |*symbol| {
const tag, pos = readEnum(Symbol.Tag, bytes, pos);
const flags, pos = readLeb(u32, bytes, pos);
symbol.* = .{
.flags = @bitCast(flags),
.name = .none,
.pointee = undefined,
};
symbol.flags.initZigSpecific(must_link, gc_sections);
.symbol_table => for (try ss.symbol_table.addManyAsSlice(gpa, count)) |*symbol| {
const tag = try br.takeEnum(Symbol.Tag, .little);
const flags: Wasm.SymbolFlags = @bitCast(try br.takeLeb128(u32));
symbol.* = .{
.flags = flags,
.name = .none,
.pointee = undefined,
};
symbol.flags.initZigSpecific(must_link, gc_sections);
switch (tag) {
.data => {
const name, pos = readBytes(bytes, pos);
const interned_name = try wasm.internString(name);
symbol.name = interned_name.toOptional();
if (symbol.flags.undefined) {
symbol.pointee = .data_import;
} else {
const segment_index, pos = readLeb(u32, bytes, pos);
const segment_offset, pos = readLeb(u32, bytes, pos);
const size, pos = readLeb(u32, bytes, pos);
try wasm.object_datas.append(gpa, .{
.segment = @enumFromInt(data_segment_start + segment_index),
.offset = segment_offset,
.size = size,
.name = interned_name,
.flags = symbol.flags,
});
symbol.pointee = .{
.data = @enumFromInt(wasm.object_datas.items.len - 1),
};
}
},
.section => {
const local_section, pos = readLeb(u32, bytes, pos);
const section: Wasm.ObjectSectionIndex = @enumFromInt(local_section_index_base + local_section);
symbol.pointee = .{ .section = section };
},
switch (tag) {
.data => {
const name = try br.take(try br.takeLeb128(u32));
const interned_name = try wasm.internString(name);
symbol.name = interned_name.toOptional();
if (symbol.flags.undefined) {
symbol.pointee = .data_import;
} else {
const segment_index = try br.takeLeb128(u32);
const segment_offset = try br.takeLeb128(u32);
const size = try br.takeLeb128(u32);
try wasm.object_datas.append(gpa, .{
.segment = @enumFromInt(data_segment_start + segment_index),
.offset = segment_offset,
.size = size,
.name = interned_name,
.flags = symbol.flags,
});
symbol.pointee = .{
.data = @enumFromInt(wasm.object_datas.items.len - 1),
};
}
},
.section => {
const local_section = try br.takeLeb128(u32);
const section: Wasm.ObjectSectionIndex = @enumFromInt(local_section_index_base + local_section);
symbol.pointee = .{ .section = section };
},
.function => {
const local_index, pos = readLeb(u32, bytes, pos);
if (symbol.flags.undefined) {
const function_import: ScratchSpace.FuncImportIndex = @enumFromInt(local_index);
symbol.pointee = .{ .function_import = function_import };
if (symbol.flags.explicit_name) {
const name, pos = readBytes(bytes, pos);
symbol.name = (try wasm.internString(name)).toOptional();
} else {
symbol.name = function_import.ptr(ss).name.toOptional();
}
} else {
symbol.pointee = .{ .function = @enumFromInt(functions_start + (local_index - ss.func_imports.items.len)) };
const name, pos = readBytes(bytes, pos);
.function => {
const local_index = try br.takeLeb128(u32);
if (symbol.flags.undefined) {
const function_import: ScratchSpace.FuncImportIndex = @enumFromInt(local_index);
symbol.pointee = .{ .function_import = function_import };
if (symbol.flags.explicit_name) {
const name = try br.take(try br.takeLeb128(u32));
symbol.name = (try wasm.internString(name)).toOptional();
}
},
.global => {
const local_index, pos = readLeb(u32, bytes, pos);
if (symbol.flags.undefined) {
const global_import: ScratchSpace.GlobalImportIndex = @enumFromInt(local_index);
symbol.pointee = .{ .global_import = global_import };
if (symbol.flags.explicit_name) {
const name, pos = readBytes(bytes, pos);
symbol.name = (try wasm.internString(name)).toOptional();
} else {
symbol.name = global_import.ptr(ss).name.toOptional();
}
} else {
symbol.pointee = .{ .global = @enumFromInt(globals_start + (local_index - ss.global_imports.items.len)) };
const name, pos = readBytes(bytes, pos);
symbol.name = (try wasm.internString(name)).toOptional();
symbol.name = function_import.ptr(ss).name.toOptional();
}
},
.table => {
const local_index, pos = readLeb(u32, bytes, pos);
if (symbol.flags.undefined) {
table_import_symbol_count += 1;
const table_import: ScratchSpace.TableImportIndex = @enumFromInt(local_index);
symbol.pointee = .{ .table_import = table_import };
if (symbol.flags.explicit_name) {
const name, pos = readBytes(bytes, pos);
symbol.name = (try wasm.internString(name)).toOptional();
} else {
symbol.name = table_import.ptr(ss).name.toOptional();
}
} else {
symbol.pointee = .{ .function = @enumFromInt(functions_start + (local_index - ss.func_imports.items.len)) };
const name = try br.take(try br.takeLeb128(u32));
symbol.name = (try wasm.internString(name)).toOptional();
}
},
.global => {
const local_index = try br.takeLeb128(u32);
if (symbol.flags.undefined) {
const global_import: ScratchSpace.GlobalImportIndex = @enumFromInt(local_index);
symbol.pointee = .{ .global_import = global_import };
if (symbol.flags.explicit_name) {
const name = try br.take(try br.takeLeb128(u32));
symbol.name = (try wasm.internString(name)).toOptional();
} else {
symbol.pointee = .{ .table = @enumFromInt(tables_start + (local_index - ss.table_imports.items.len)) };
const name, pos = readBytes(bytes, pos);
symbol.name = (try wasm.internString(name)).toOptional();
symbol.name = global_import.ptr(ss).name.toOptional();
}
},
else => {
log.debug("unrecognized symbol type tag: {x}", .{@intFromEnum(tag)});
return error.UnrecognizedSymbolType;
},
}
} else {
symbol.pointee = .{ .global = @enumFromInt(globals_start + (local_index - ss.global_imports.items.len)) };
const name = try br.take(try br.takeLeb128(u32));
symbol.name = (try wasm.internString(name)).toOptional();
}
},
.table => {
const local_index = try br.takeLeb128(u32);
if (symbol.flags.undefined) {
table_import_symbol_count += 1;
const table_import: ScratchSpace.TableImportIndex = @enumFromInt(local_index);
symbol.pointee = .{ .table_import = table_import };
if (symbol.flags.explicit_name) {
const name = try br.take(try br.takeLeb128(u32));
symbol.name = (try wasm.internString(name)).toOptional();
} else {
symbol.name = table_import.ptr(ss).name.toOptional();
}
} else {
symbol.pointee = .{ .table = @enumFromInt(tables_start + (local_index - ss.table_imports.items.len)) };
const name = try br.take(try br.takeLeb128(u32));
symbol.name = (try wasm.internString(name)).toOptional();
}
},
else => {
log.debug("unrecognized symbol type tag: {x}", .{@intFromEnum(tag)});
return error.UnrecognizedSymbolType;
},
}
},
}
@ -504,8 +488,8 @@ pub fn parse(
// which section they apply to, and must be sequenced in
// the module after that section."
// "Relocation sections can only target code, data and custom sections."
const local_section, pos = readLeb(u32, bytes, pos);
const count, pos = readLeb(u32, bytes, pos);
const local_section = try br.takeLeb128(u32);
const count = try br.takeLeb128(u32);
const section: Wasm.ObjectSectionIndex = @enumFromInt(local_section_index_base + local_section);
log.debug("found {d} relocations for section={d}", .{ count, section });
@ -513,10 +497,9 @@ pub fn parse(
var prev_offset: u32 = 0;
try wasm.object_relocations.ensureUnusedCapacity(gpa, count);
for (0..count) |_| {
const tag: RelocationType = @enumFromInt(bytes[pos]);
pos += 1;
const offset, pos = readLeb(u32, bytes, pos);
const index, pos = readLeb(u32, bytes, pos);
const tag = try br.takeEnum(RelocationType, .little);
const offset = try br.takeLeb128(u32);
const index = try br.takeLeb128(u32);
if (offset < prev_offset)
return diags.failParse(path, "relocation entries not sorted by offset", .{});
@ -537,7 +520,7 @@ pub fn parse(
.memory_addr_locrel_i32,
.memory_addr_tls_sleb64,
=> {
const addend: i32, pos = readLeb(i32, bytes, pos);
const addend = try br.takeLeb128(i32);
wasm.object_relocations.appendAssumeCapacity(switch (sym.pointee) {
.data => |data| .{
.tag = .fromType(tag),
@ -555,7 +538,7 @@ pub fn parse(
});
},
.function_offset_i32, .function_offset_i64 => {
const addend: i32, pos = readLeb(i32, bytes, pos);
const addend = try br.takeLeb128(i32);
wasm.object_relocations.appendAssumeCapacity(switch (sym.pointee) {
.function => .{
.tag = .fromType(tag),
@ -573,7 +556,7 @@ pub fn parse(
});
},
.section_offset_i32 => {
const addend: i32, pos = readLeb(i32, bytes, pos);
const addend = try br.takeLeb128(i32);
wasm.object_relocations.appendAssumeCapacity(.{
.tag = .section_offset_i32,
.offset = offset,
@ -658,10 +641,9 @@ pub fn parse(
.len = count,
});
} else if (std.mem.eql(u8, section_name, "target_features")) {
opt_features, pos = try parseFeatures(wasm, bytes, pos, path);
opt_features = try parseFeatures(wasm, br, path);
} else if (std.mem.startsWith(u8, section_name, ".debug")) {
const debug_content = bytes[pos..section_end];
pos = section_end;
const debug_content = try br.take(len);
const data_off: u32 = @intCast(wasm.string_bytes.items.len);
try wasm.string_bytes.appendSlice(gpa, debug_content);
@ -669,23 +651,20 @@ pub fn parse(
try wasm.object_custom_segments.put(gpa, section_index, .{
.payload = .{
.off = @enumFromInt(data_off),
.len = @intCast(debug_content.len),
.len = @intCast(len),
},
.flags = .{},
.section_name = try wasm.internString(section_name),
});
} else {
pos = section_end;
}
} else br.seek = section_end;
},
.type => {
const func_types_len, pos = readLeb(u32, bytes, pos);
const func_types_len = try br.takeLeb128(u32);
for (try ss.func_types.addManyAsSlice(gpa, func_types_len)) |*func_type| {
if (bytes[pos] != std.wasm.function_type) return error.ExpectedFuncType;
pos += 1;
if (try br.takeByte() != std.wasm.function_type) return error.ExpectedFuncType;
const params, pos = readBytes(bytes, pos);
const returns, pos = readBytes(bytes, pos);
const params = try br.take(try br.takeLeb128(u32));
const returns = try br.take(try br.takeLeb128(u32));
func_type.* = try wasm.addFuncType(.{
.params = .fromString(try wasm.internString(params)),
.returns = .fromString(try wasm.internString(returns)),
@ -693,16 +672,16 @@ pub fn parse(
}
},
.import => {
const imports_len, pos = readLeb(u32, bytes, pos);
const imports_len = try br.takeLeb128(u32);
for (0..imports_len) |_| {
const module_name, pos = readBytes(bytes, pos);
const name, pos = readBytes(bytes, pos);
const kind, pos = readEnum(std.wasm.ExternalKind, bytes, pos);
const module_name = try br.take(try br.takeLeb128(u32));
const name = try br.take(try br.takeLeb128(u32));
const kind = try br.takeEnum(std.wasm.ExternalKind, .little);
const interned_module_name = try wasm.internString(module_name);
const interned_name = try wasm.internString(name);
switch (kind) {
.function => {
const function, pos = readLeb(u32, bytes, pos);
const function = try br.takeLeb128(u32);
try ss.func_imports.append(gpa, .{
.module_name = interned_module_name,
.name = interned_name,
@ -710,7 +689,7 @@ pub fn parse(
});
},
.memory => {
const limits, pos = readLimits(bytes, pos);
const limits = try readLimits(br);
const gop = try wasm.object_memory_imports.getOrPut(gpa, interned_name);
if (gop.found_existing) {
if (gop.value_ptr.module_name != interned_module_name) {
@ -736,9 +715,12 @@ pub fn parse(
}
},
.global => {
const valtype, pos = readEnum(std.wasm.Valtype, bytes, pos);
const mutable = bytes[pos] == 0x01;
pos += 1;
const valtype = try br.takeEnum(std.wasm.Valtype, .little);
const mutable = switch (try br.takeByte()) {
0 => false,
1 => true,
else => return error.InvalidMutability,
};
try ss.global_imports.append(gpa, .{
.name = interned_name,
.valtype = valtype,
@ -747,8 +729,8 @@ pub fn parse(
});
},
.table => {
const ref_type, pos = readEnum(std.wasm.RefType, bytes, pos);
const limits, pos = readLimits(bytes, pos);
const ref_type = try br.takeEnum(std.wasm.RefType, .little);
const limits = try readLimits(br);
try ss.table_imports.append(gpa, .{
.name = interned_name,
.module_name = interned_module_name,
@ -763,17 +745,16 @@ pub fn parse(
}
},
.function => {
const functions_len, pos = readLeb(u32, bytes, pos);
const functions_len = try br.takeLeb128(u32);
for (try ss.func_type_indexes.addManyAsSlice(gpa, functions_len)) |*func_type_index| {
const i, pos = readLeb(u32, bytes, pos);
func_type_index.* = @enumFromInt(i);
func_type_index.* = @enumFromInt(try br.takeLeb128(u32));
}
},
.table => {
const tables_len, pos = readLeb(u32, bytes, pos);
const tables_len = try br.takeLeb128(u32);
for (try wasm.object_tables.addManyAsSlice(gpa, tables_len)) |*table| {
const ref_type, pos = readEnum(std.wasm.RefType, bytes, pos);
const limits, pos = readLimits(bytes, pos);
const ref_type = try br.takeEnum(std.wasm.RefType, .little);
const limits = try readLimits(br);
table.* = .{
.name = .none,
.module_name = .none,
@ -788,9 +769,9 @@ pub fn parse(
}
},
.memory => {
const memories_len, pos = readLeb(u32, bytes, pos);
const memories_len = try br.takeLeb128(u32);
for (try wasm.object_memories.addManyAsSlice(gpa, memories_len)) |*memory| {
const limits, pos = readLimits(bytes, pos);
const limits = try readLimits(br);
memory.* = .{
.name = .none,
.flags = .{
@ -807,14 +788,17 @@ pub fn parse(
return diags.failParse(path, "object has more than one global section", .{});
global_section_index = section_index;
const section_start = pos;
const globals_len, pos = readLeb(u32, bytes, pos);
const section_start = br.seek;
const globals_len = try br.takeLeb128(u32);
for (try wasm.object_globals.addManyAsSlice(gpa, globals_len)) |*global| {
const valtype, pos = readEnum(std.wasm.Valtype, bytes, pos);
const mutable = bytes[pos] == 0x01;
pos += 1;
const init_start = pos;
const expr, pos = try readInit(wasm, bytes, pos);
const valtype = try br.takeEnum(std.wasm.Valtype, .little);
const mutable = switch (try br.takeByte()) {
0 => false,
1 => true,
else => return error.InvalidMutability,
};
const init_start = br.seek;
const expr = try readInit(wasm, br);
global.* = .{
.name = .none,
.flags = .{
@ -826,20 +810,19 @@ pub fn parse(
.expr = expr,
.object_index = object_index,
.offset = @intCast(init_start - section_start),
.size = @intCast(pos - init_start),
.size = @intCast(br.seek - init_start),
};
}
},
.@"export" => {
const exports_len, pos = readLeb(u32, bytes, pos);
const exports_len = try br.takeLeb128(u32);
// Read into scratch space, and then later add this data as if
// it were extra symbol table entries, but allow merging with
// existing symbol table data if the name matches.
for (try ss.exports.addManyAsSlice(gpa, exports_len)) |*exp| {
const name, pos = readBytes(bytes, pos);
const kind: std.wasm.ExternalKind = @enumFromInt(bytes[pos]);
pos += 1;
const index, pos = readLeb(u32, bytes, pos);
const name = try br.take(try br.takeLeb128(u32));
const kind = try br.takeEnum(std.wasm.ExternalKind, .little);
const index = try br.takeLeb128(u32);
exp.* = .{
.name = try wasm.internString(name),
.pointee = switch (kind) {
@ -852,25 +835,24 @@ pub fn parse(
}
},
.start => {
const index, pos = readLeb(u32, bytes, pos);
const index = try br.takeLeb128(u32);
start_function = @enumFromInt(functions_start + index);
},
.element => {
log.warn("unimplemented: element section in {f} {?s}", .{ path, archive_member_name });
pos = section_end;
br.seek = section_end;
},
.code => {
if (code_section_index != null)
return diags.failParse(path, "object has more than one code section", .{});
code_section_index = section_index;
const start = pos;
const count, pos = readLeb(u32, bytes, pos);
const start = br.seek;
const count = try br.takeLeb128(u32);
for (try wasm.object_functions.addManyAsSlice(gpa, count)) |*elem| {
const code_len, pos = readLeb(u32, bytes, pos);
const offset: u32 = @intCast(pos - start);
const payload = try wasm.addRelocatableDataPayload(bytes[pos..][0..code_len]);
pos += code_len;
const code_len = try br.takeLeb128(u32);
const offset: u32 = @intCast(br.seek - start);
const payload = try wasm.addRelocatableDataPayload(try br.take(code_len));
elem.* = .{
.flags = .{}, // populated from symbol table
.name = .none, // populated from symbol table
@ -886,20 +868,19 @@ pub fn parse(
return diags.failParse(path, "object has more than one data section", .{});
data_section_index = section_index;
const section_start = pos;
const count, pos = readLeb(u32, bytes, pos);
const section_start = br.seek;
const count = try br.takeLeb128(u32);
for (try wasm.object_data_segments.addManyAsSlice(gpa, count)) |*elem| {
const flags, pos = readEnum(DataSegmentFlags, bytes, pos);
const flags: DataSegmentFlags = @enumFromInt(try br.takeLeb128(u32));
if (flags == .active_memidx) {
const memidx, pos = readLeb(u32, bytes, pos);
const memidx = try br.takeLeb128(u32);
if (memidx != 0) return diags.failParse(path, "data section uses mem index {d}", .{memidx});
}
//const expr, pos = if (flags != .passive) try readInit(wasm, bytes, pos) else .{ .none, pos };
if (flags != .passive) pos = try skipInit(bytes, pos);
const data_len, pos = readLeb(u32, bytes, pos);
const segment_start = pos;
const payload = try wasm.addRelocatableDataPayload(bytes[pos..][0..data_len]);
pos += data_len;
//const expr = if (flags != .passive) try readInit(wasm, br) else .none;
if (flags != .passive) try skipInit(br);
const data_len = try br.takeLeb128(u32);
const segment_start = br.seek;
const payload = try wasm.addRelocatableDataPayload(try br.take(data_len));
elem.* = .{
.payload = payload,
.name = .none, // Populated from segment_info
@ -911,10 +892,10 @@ pub fn parse(
};
}
},
else => pos = section_end,
else => br.seek = section_end,
}
if (pos != section_end) return error.MalformedSection;
}
if (br.seek != section_end) return error.MalformedSection;
} else |_| {}
if (!saw_linking_section) return error.MissingLinkingSection;
const cpu = comp.root_mod.resolved_target.result.cpu;
@ -1422,27 +1403,27 @@ pub fn parse(
/// Based on the "features" custom section, parses it into a list of
/// features that tell the linker what features were enabled and may be mandatory
/// to be able to link.
fn parseFeatures(
wasm: *Wasm,
bytes: []const u8,
start_pos: usize,
path: Path,
) error{ OutOfMemory, LinkFailure }!struct { Wasm.Feature.Set, usize } {
fn parseFeatures(wasm: *Wasm, reader: *Reader, path: Path) error{ OutOfMemory, LinkFailure }!Wasm.Feature.Set {
const gpa = wasm.base.comp.gpa;
const diags = &wasm.base.comp.link_diags;
const features_len, var pos = readLeb(u32, bytes, start_pos);
const features_len = reader.takeLeb128(u32) catch |err|
return diags.failParse(path, "invalid features length: {t}", .{err});
// This temporary allocation could be avoided by using the string_bytes buffer as a scratch space.
const feature_buffer = try gpa.alloc(Wasm.Feature, features_len);
defer gpa.free(feature_buffer);
for (feature_buffer) |*feature| {
const prefix: Wasm.Feature.Prefix = switch (bytes[pos]) {
const prefix: Wasm.Feature.Prefix = switch (reader.takeByte() catch |err| {
return diags.failParse(path, "invalid feature prefix: {t}", .{err});
}) {
'-' => .@"-",
'+' => .@"+",
'=' => .@"=",
else => |b| return diags.failParse(path, "invalid feature prefix: 0x{x}", .{b}),
};
pos += 1;
const name, pos = readBytes(bytes, pos);
const name_len = reader.takeLeb128(u32) catch |err|
return diags.failParse(path, "bad feature name length: {t}", .{err});
const name = reader.take(name_len) catch |err|
return diags.failParse(path, "bad feature name: {t}", .{err});
const tag = std.meta.stringToEnum(Wasm.Feature.Tag, name) orelse {
return diags.failParse(path, "unrecognized wasm feature in object: {s}", .{name});
};
@ -1453,68 +1434,34 @@ fn parseFeatures(
}
std.mem.sortUnstable(Wasm.Feature, feature_buffer, {}, Wasm.Feature.lessThan);
return .fromString(try wasm.internString(@ptrCast(feature_buffer)));
}
fn readLimits(reader: *Reader) !std.wasm.Limits {
const flags: std.wasm.Limits.Flags = @bitCast(try reader.takeByte());
const min = try reader.takeLeb128(u32);
const max = if (flags.has_max) try reader.takeLeb128(u32) else 0;
return .{
.fromString(try wasm.internString(@ptrCast(feature_buffer))),
pos,
};
}
fn readLeb(comptime T: type, bytes: []const u8, pos: usize) struct { T, usize } {
var fbr = std.io.fixedBufferStream(bytes[pos..]);
return .{
switch (@typeInfo(T).int.signedness) {
.signed => std.leb.readIleb128(T, fbr.reader()) catch unreachable,
.unsigned => std.leb.readUleb128(T, fbr.reader()) catch unreachable,
},
pos + fbr.pos,
};
}
fn readBytes(bytes: []const u8, start_pos: usize) struct { []const u8, usize } {
const len, const pos = readLeb(u32, bytes, start_pos);
return .{
bytes[pos..][0..len],
pos + len,
};
}
fn readEnum(comptime T: type, bytes: []const u8, pos: usize) struct { T, usize } {
const Tag = @typeInfo(T).@"enum".tag_type;
const int, const new_pos = readLeb(Tag, bytes, pos);
return .{ @enumFromInt(int), new_pos };
}
fn readLimits(bytes: []const u8, start_pos: usize) struct { std.wasm.Limits, usize } {
const flags: std.wasm.Limits.Flags = @bitCast(bytes[start_pos]);
const min, const max_pos = readLeb(u32, bytes, start_pos + 1);
const max, const end_pos = if (flags.has_max) readLeb(u32, bytes, max_pos) else .{ 0, max_pos };
return .{ .{
.flags = flags,
.min = min,
.max = max,
}, end_pos };
}
fn readInit(wasm: *Wasm, bytes: []const u8, pos: usize) !struct { Wasm.Expr, usize } {
const end_pos = try skipInit(bytes, pos); // one after the end opcode
return .{ try wasm.addExpr(bytes[pos..end_pos]), end_pos };
}
pub fn exprEndPos(bytes: []const u8, pos: usize) error{InvalidInitOpcode}!usize {
const opcode = bytes[pos];
return switch (@as(std.wasm.Opcode, @enumFromInt(opcode))) {
.i32_const => readLeb(i32, bytes, pos + 1)[1],
.i64_const => readLeb(i64, bytes, pos + 1)[1],
.f32_const => pos + 5,
.f64_const => pos + 9,
.global_get => readLeb(u32, bytes, pos + 1)[1],
else => return error.InvalidInitOpcode,
};
}
fn skipInit(bytes: []const u8, pos: usize) !usize {
const end_pos = try exprEndPos(bytes, pos);
const op, const final_pos = readEnum(std.wasm.Opcode, bytes, end_pos);
if (op != .end) return error.InitExprMissingEnd;
return final_pos;
fn readInit(wasm: *Wasm, reader: *Reader) !Wasm.Expr {
const start = reader.seek;
try skipInit(reader); // one after the end opcode
return wasm.addExpr(reader.buffered()[start..reader.seek]);
}
pub fn skipInit(reader: *Reader) !void {
switch (try reader.takeEnumNonexhaustive(std.wasm.Opcode, .little)) {
.i32_const => _ = try reader.takeLeb128(i32),
.i64_const => _ = try reader.takeLeb128(i64),
.f32_const => try reader.discardAll(5),
.f64_const => try reader.discardAll(9),
.global_get => _ = try reader.takeLeb128(u32),
else => return error.InvalidInitOpcode,
}
if (try reader.takeEnum(std.wasm.Opcode, .little) != .end) return error.InitExprMissingEnd;
}