Files
Odin/core/rexcode/ir/wasm/encoder.odin
2026-07-14 11:29:49 +01:00

224 lines
6.1 KiB
Odin

// rexcode · Brendan Punsky (dotbmp@github), original author
// Ginger Bill (gingerBill@github)
package rexcode_wasm
import "core:math/bits"
// =============================================================================
// SECTION: Encoder (Module -> WASM instruction byte stream)
// =============================================================================
//
// Variable-length, byte-oriented, LEB128-heavy. Encoding is a single forward
// pass: each operation writes its opcode (a byte, or a prefix byte plus an
// unsigned-LEB sub-opcode) followed by its immediates, advancing a byte cursor.
//
// WASM has no PC-relative branches (control flow uses structured label depths),
// so there is no second resolution pass -- which is exactly why the ir verbs
// (docs/ir_design.md §4) drop the ISA `label_defs`/`resolve`/`base_address`.
// Relocations *are* produced, for symbolic index references (see op_label), and
// returned for a linker to patch; symbolic indices are laid down as fixed-width
// 5-byte LEB placeholders so the patched value always fits.
//
// The reusable core is `encode_ops` (an operation stream = a WASM `expr`). The
// container `encode` verb (write.odin) wraps those bodies in the CODE-section
// framing and emits the surrounding sections, producing a whole `.wasm` file;
// this file owns only the instruction-stream level.
MAX_OPCODE_SIZE :: 3 // prefix byte + two-byte unsigned-LEB sub-opcode (SIMD reaches 0x113)
@(require_results)
encode_max_code_size :: #force_inline proc "contextless" (n: int) -> int {
// Worst case per op without a br_table: a 3-byte opcode plus the largest
// single immediate, v128.const's 16 raw bytes. br_table is unbounded in its
// target count; callers encoding those should size from the target totals.
return n * 24
}
@(require_results)
encode_max_relocation_count :: #force_inline proc "contextless" (n: int) -> int {
return n
}
// encode_ops: the reusable instruction-stream encoder (a WASM `expr`).
encode_ops :: proc(ops: []Operation, code: []u8, relocs: ^[dynamic]Relocation, errors: ^[dynamic]Error) -> (byte_count: u32, ok: bool) {
errors_start := u32(len(errors))
for &op, i in ops {
n := encode_operation(&op, byte_count, u16(i), code, relocs, errors) or_return
byte_count += n
}
ok = u32(len(errors)) == errors_start
return
}
encode_operation :: proc(
op: ^Operation,
pc: u32,
op_index: u16,
code: []u8,
relocs: ^[dynamic]Relocation,
errors: ^[dynamic]Error,
) -> (size: u32, ok: bool) {
opcode := Opcode(op.opcode)
if opcode == .INVALID {
append(errors, Error{location = u32(op_index), code = .INVALID_OPCODE})
return
}
form := encoding_form(opcode)
need := encoded_size(op, form)
if pc + need > u32(len(code)) {
append(errors, Error{location = u32(op_index), code = .BUFFER_OVERFLOW})
return
}
off := pc
// Opcode (and prefix sub-opcode).
if form.prefix == PREFIX_NONE {
code[off] = u8(form.opcode)
off += 1
} else {
code[off] = form.prefix
off += 1
write_uleb(code, &off, u64(form.opcode))
}
// Immediates, walked in declaration order with an operand cursor.
opi := 0
for k in form.imm {
switch k {
case .NONE:
// nothing
case .BLOCKTYPE, .I32, .I64:
write_sleb(code, &off, op.operands[opi].imm)
opi += 1
case .F32:
write_u32_block(code, &off, u32(op.operands[opi].imm))
opi += 1
case .F64:
write_u64_block(code, &off, u64(op.operands[opi].imm))
opi += 1
case .IDX:
o := op.operands[opi]
if operand_symbolic(o) {
append(relocs, Relocation{
offset = off, label_id = operand_index(o), addend = 0,
type = reloc_type_for(operand_index_kind(o)), size = 5, inst_idx = op_index,
})
write_uleb_padded5(code, &off, u64(operand_index(o)))
} else {
write_uleb(code, &off, u64(operand_index(o)))
}
opi += 1
case .MEMARG:
ma := operand_memarg(op.operands[opi])
// NOTE(bill): stored as log2 even though the spec text reads otherwise.
align := bits.log2(u64(ma.align))
write_uleb(code, &off, align)
write_uleb(code, &off, u64(ma.offset))
opi += 1
case .REFTYPE:
code[off] = u8(op.operands[opi].imm)
off += 1
opi += 1
case .BR_TABLE:
// operands = [default, case0, case1, ...] -- every entry a label depth.
cases := op.operands[opi + 1:]
write_uleb(code, &off, u64(len(cases)))
for c in cases {
write_uleb(code, &off, u64(operand_index(c)))
}
write_uleb(code, &off, u64(operand_index(op.operands[opi]))) // default
opi = len(op.operands)
case .ZERO_BYTE:
code[off] = 0x00
off += 1
case .LANE:
code[off] = u8(op.operands[opi].imm)
off += 1
opi += 1
case .LANES16:
bytes := operand_v128(op.operands[opi], op.operands[opi + 1])
for bb in bytes {
code[off] = bb
off += 1
}
opi += 2
}
}
return off - pc, true
}
@(private="file")
encoded_size :: proc(op: ^Operation, form: ^Encoding) -> u32 {
size: u32 = 1
if form.prefix != PREFIX_NONE {
size += uleb_size(u64(form.opcode))
}
opi := 0
for k in form.imm {
switch k {
case .NONE:
case .BLOCKTYPE, .I32, .I64:
size += sleb_size(op.operands[opi].imm)
opi += 1
case .F32:
size += 4
opi += 1
case .F64:
size += 8
opi += 1
case .IDX:
o := op.operands[opi]
size += operand_symbolic(o) ? 5 : uleb_size(u64(operand_index(o)))
opi += 1
case .MEMARG:
ma := operand_memarg(op.operands[opi])
size += uleb_size(bits.log2(u64(ma.align))) + uleb_size(u64(ma.offset))
opi += 1
case .REFTYPE:
size += 1
opi += 1
case .BR_TABLE:
cases := op.operands[opi + 1:]
size += uleb_size(u64(len(cases)))
for c in cases {
size += uleb_size(u64(operand_index(c)))
}
size += uleb_size(u64(operand_index(op.operands[opi])))
opi = len(op.operands)
case .ZERO_BYTE:
size += 1
case .LANE:
size += 1
opi += 1
case .LANES16:
size += 16
opi += 2
}
}
return size
}
@(private="file")
reloc_type_for :: #force_inline proc "contextless" (k: Index_Kind) -> Relocation_Type {
#partial switch k {
case .FUNC: return .FUNCTION_INDEX_LEB
case .TYPE: return .TYPE_INDEX_LEB
case .GLOBAL: return .GLOBAL_INDEX_LEB
case .TABLE: return .TABLE_NUMBER_LEB
}
return .FUNCTION_INDEX_LEB
}