perf: 字符串域加模板折叠为单字符串常量

This commit is contained in:
2026-10-12 13:15:05 +00:00
parent 0aa7bf83c1
commit 35efd73ea0
7 changed files with 257 additions and 31 deletions
Generated
+20 -20
View File
@@ -1160,7 +1160,7 @@ dependencies = [
[[package]]
name = "oxide_builtins"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"chrono",
"fast-float",
@@ -1185,7 +1185,7 @@ dependencies = [
[[package]]
name = "oxide_bytecode"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"num-bigint",
"oxide_log",
@@ -1193,7 +1193,7 @@ dependencies = [
[[package]]
name = "oxide_cfg"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_emit",
@@ -1204,7 +1204,7 @@ dependencies = [
[[package]]
name = "oxide_cli"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"ansi_term",
"bumpalo",
@@ -1228,7 +1228,7 @@ dependencies = [
[[package]]
name = "oxide_code_cache"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"lru",
"oxide_bytecode",
@@ -1237,7 +1237,7 @@ dependencies = [
[[package]]
name = "oxide_compiler"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1257,7 +1257,7 @@ dependencies = [
[[package]]
name = "oxide_dce"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1271,7 +1271,7 @@ dependencies = [
[[package]]
name = "oxide_emit"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"num-bigint",
"oxc_allocator",
@@ -1287,7 +1287,7 @@ dependencies = [
[[package]]
name = "oxide_engine"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_compiler",
@@ -1299,7 +1299,7 @@ dependencies = [
[[package]]
name = "oxide_ir"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_log",
@@ -1308,7 +1308,7 @@ dependencies = [
[[package]]
name = "oxide_kernel"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"bumpalo",
"dashmap",
@@ -1322,7 +1322,7 @@ dependencies = [
[[package]]
name = "oxide_liveness"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1334,7 +1334,7 @@ dependencies = [
[[package]]
name = "oxide_log"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"tracing",
"tracing-appender",
@@ -1343,11 +1343,11 @@ dependencies = [
[[package]]
name = "oxide_net"
version = "0.16.15"
version = "0.16.16"
[[package]]
name = "oxide_parser"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxc_allocator",
"oxc_ast",
@@ -1362,7 +1362,7 @@ dependencies = [
[[package]]
name = "oxide_regalloc"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1378,7 +1378,7 @@ dependencies = [
[[package]]
name = "oxide_runtime_api"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"num-bigint",
"num-traits",
@@ -1391,7 +1391,7 @@ dependencies = [
[[package]]
name = "oxide_test262"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"oxide_bytecode",
"oxide_compiler",
@@ -1411,7 +1411,7 @@ dependencies = [
[[package]]
name = "oxide_types"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"bumpalo",
"dashmap",
@@ -1424,7 +1424,7 @@ dependencies = [
[[package]]
name = "oxide_vm"
version = "0.16.15"
version = "0.16.16"
dependencies = [
"bumpalo",
"chrono",
+1 -1
View File
@@ -24,7 +24,7 @@ members = [
]
[workspace.package]
version = "0.16.15"
version = "0.16.16"
edition = "2021"
license = "MIT"
rust-version = "1.80"
+99 -4
View File
@@ -1,18 +1,29 @@
//! 双常量折叠:两操作数均为字面量且通过语义守卫时,编译期折成单常量,
//! 常量折叠:操作数均为字面量且通过语义守卫时,编译期折成单常量,
//! 发射单条 `LOAD_CONST`。折值编码逐运算符镜像运行时指令的产物编码,
//! 保证物化后与运行时执行结果逐字节一致。
//!
//! 三条语义守卫:任一操作数为 BigInt 字面量不折(BigInt 混合运行期抛
//! 两个域并列:数值域(`literal_constant`/`fold_binary`,两操作数形态,
//! 守卫排除字符串与 BigInt)与字符串域(`literal_units`/`fold_string_chain`/
//! `fold_template`,Addition 链与全字面量模板,单元口径与运行期
//! `push_units_to` 逐段一致)。
//!
//! 数值域语义守卫:任一操作数为 BigInt 字面量不折(BigInt 混合运行期抛
//! TypeError,BigInt 与 BigInt 有独立指令族);`DIV`/`MOD` 除数为零不折、
//! `EXP` 指数为负不折(保持运行时指令为除零/负指数编码的唯一来源)。
//! 字符串字面量不折(字符串域是独立任务)。
//! 字符串域语义守卫:Addition 链须至少一个字符串字面量(全数字链落回
//! 数值域与 `CONCAT_N`,两域零重叠);模板须全部 cooked 为 Some。
use oxide_bytecode::module::Constant;
use oxide_parser::{BinaryOperator, Expression, UnaryOperator};
use oxide_runtime_api::{abstract_eq_primitives, relational_compare, strict_equality, to_int32, to_number, to_uint32};
use oxide_runtime_api::{
abstract_eq_primitives, bigint_to_string, js_number_to_string, relational_compare, strict_equality, to_int32,
to_number, to_uint32,
};
use oxide_types::string_forge::{decode_key, encode_key};
use oxide_types::value::JsValue;
use crate::is_int_literal;
use crate::shared::string_pool::pool_key_marker;
/// 字面量表达式提取常量:剥括号与一元负号后,数字(`is_int_literal` 分派
/// `Int`/`Number`)、布尔、null 三类;BigInt 字面量、字符串字面量、RegExp 与
@@ -186,3 +197,87 @@ fn relational(l: &Constant, r: &Constant, swap: bool, negate: bool) -> Option<Co
let cmp = relational_compare(a, b).unwrap_or(false);
Some(Constant::Boolean(if negate { !cmp } else { cmp }))
}
/// 可折叠字面量的单元序列(剥括号与一元负号链):字符串字面量经
/// `pool_key_marker` 加 `decode_key` 还原单元(与字面量发射器同一池键函数);
/// 数字字面量按 `is_int_literal` 分派 i32 itoa 或 `js_number_to_string` 后
/// UTF-16 编码;BigInt 字面量解析后 `bigint_to_string`(i128 快路径失败回退
/// 十进制 `parse_bytes`,解析失败返回 `None` 不折);布尔与 null 为
/// "true"/"false"/"null"。
///
/// 其余形态(标识符/调用/RegExp/嵌套二元等)返回 `None`。
pub(crate) fn literal_units(expr: &Expression) -> Option<Vec<u16>> {
let (inner, negated) = peel_negation(expr);
match inner {
Expression::StringLiteral(s) if !negated => Some(decode_key(&pool_key_marker(&s.value, s.lone_surrogates))),
Expression::NumericLiteral(n) => {
let value = if negated { -n.value } else { n.value };
let text = if is_int_literal(value) {
(value as i32).to_string()
} else {
js_number_to_string(value)
};
Some(text.encode_utf16().collect())
}
Expression::BigIntLiteral(b) => {
let value = if let Ok(v) = b.value.parse::<i128>() {
num_bigint::BigInt::from(v)
} else {
num_bigint::BigInt::parse_bytes(b.value.as_bytes(), 10)?
};
let value = if negated { -value } else { value };
Some(bigint_to_string(&value).encode_utf16().collect())
}
Expression::BooleanLiteral(b) if !negated => {
let text = if b.value { "true" } else { "false" };
Some(text.encode_utf16().collect())
}
Expression::NullLiteral(_) if !negated => Some("null".encode_utf16().collect()),
_ => None,
}
}
/// 判定操作数是否为字符串字面量(剥括号):规范判定整条 `+` 链为字符串
/// 拼接的条件。
fn is_string_literal(expr: &Expression) -> bool {
let mut e = expr;
while let Expression::ParenthesizedExpression(p) = e {
e = &p.expression;
}
matches!(e, Expression::StringLiteral(_))
}
/// Addition 链字符串域折叠:全操作数经 `literal_units` 可定值且至少一个是
/// 字符串字面量时,拼接各段单元经 `encode_key` 回池键形态,返回
/// `Constant::String`;全数字链(无字符串字面量)返回 `None` 落回 `CONCAT_N`。
pub(crate) fn fold_string_chain(operands: &[&Expression]) -> Option<Constant> {
// 字符串域守卫置于首行:至少一个字符串字面量,全数字链留给数值域。
if !operands.iter().any(|op| is_string_literal(op)) {
return None;
}
let mut units: Vec<u16> = Vec::new();
for op in operands {
units.extend(literal_units(op)?);
}
Some(Constant::String(encode_key(&units)))
}
/// 模板字面量折叠:全部 quasi 的 cooked 为 Some 且全部内插表达式经
/// `literal_units` 可定值时,按 quasi 与表达式交替序拼接单元经 `encode_key`
/// 回池键形态,返回 `Constant::String`;任一 cooked 为 None(非法转义,
/// 解析期 SyntaxError 的防御路径)返回 `None` 落回 `TEMPLATE_STR`。
pub(crate) fn fold_template(tl: &oxide_parser::TemplateLiteral) -> Option<Constant> {
// cooked 守卫置于首行:任一段为 None 时整模板不折。
if !tl.quasis.iter().all(|q| q.value.cooked.is_some()) {
return None;
}
let mut units: Vec<u16> = Vec::new();
for (i, q) in tl.quasis.iter().enumerate() {
let cooked = q.value.cooked.as_ref().expect("守卫已确认全部 cooked 为 Some");
units.extend(decode_key(&pool_key_marker(cooked, q.lone_surrogates)));
if let Some(expr) = tl.expressions.get(i) {
units.extend(literal_units(expr)?);
}
}
Some(Constant::String(encode_key(&units)))
}
+9 -1
View File
@@ -2,7 +2,7 @@
//! 逻辑与 `??` 用短路跳转,复合求值利用 `is_side_effect_free` 优化。
//! 函数:`emit_operator` 及各类 `emit_*_expression`。
mod fold;
pub(crate) mod fold;
use crate::{is_side_effect_free, BinaryOperator, CompileCtx, Emitter};
use oxide_bytecode::module::Constant;
@@ -50,6 +50,14 @@ impl Emitter {
let mut operands = Vec::new();
collect_add_operands(&bin.left, &mut operands);
operands.push(&bin.right);
// 字符串域链折叠:全操作数为可折叠字面量且至少一个字符串字面量时,
// 编译期折单字符串常量(单元口径见 fold.rs),发射单条 LOAD_CONST。
if let Some(folded) = fold::fold_string_chain(&operands) {
let idx = ctx.add_constant(folded);
let r = ctx.alloc_reg();
ctx.inst(Inst::load_const(Operand::Reg(r), idx));
return Ok(r);
}
// 上限防寄存器压力:CONCAT_N 单点读全部操作数(须同时存活),超长链回退
// 左结合 ADD(寄存器随链复用,纯表达式寄存器占用有界)。
if operands.len() >= 3 && operands.len() <= MAX_CONCAT_N_OPERANDS {
@@ -10,6 +10,14 @@ impl Emitter {
pub(crate) fn emit_template_literal_expression(
&self, tl: &oxide_parser::TemplateLiteral, ctx: &mut CompileCtx,
) -> Result<u32, String> {
// 全字面量模板折叠:全部内插表达式为可折叠字面量且全部 cooked 为 Some 时,
// 编译期折单字符串常量,发射单条 LOAD_CONST(单元口径见 fold.rs)。
if let Some(folded) = crate::expr::operator::fold::fold_template(tl) {
let idx = ctx.add_constant(folded);
let r = ctx.alloc_reg();
ctx.inst(Inst::load_const(Operand::Reg(r), idx));
return Ok(r);
}
let r = ctx.alloc_reg();
let quasis = &tl.quasis;
let expressions = &tl.expressions;
+116 -2
View File
@@ -265,10 +265,11 @@ fn destructured_rest_and_holes() {
}
/// 面 6:模板字面量的 TEMPLATE_STR 面与 quasis 池键(键经 string_forge 编码,
/// 纯 ASCII 键按值还原即原形态)。
/// 纯 ASCII 键按值还原即原形态)。全字面量模板已折单常量(见字符串域折叠
/// 测试),此处用非字面量内插表达式保持 TEMPLATE_STR 路径。
#[test]
fn template_literal_pool_keys() {
let src = "\"use strict\";\nconst t = `a${1}b`;\nconst tag = (s) => s[0];\ntag`x${2}y`";
let src = "\"use strict\";\nvar x = 1;\nconst t = `a${x}b`;\nconst tag = (s) => s[0];\ntag`x${2}y`";
let ir = emit_ir(src);
assert!(count_op(&ir.insts, OpCode::TEMPLATE_STR) >= 1);
let pool = pool_strings(&ir);
@@ -422,3 +423,116 @@ fn binary_literal_fold_entry() {
assert!(ir.constants.contains(&Constant::Number(6.0)), "常量池应含乘法折值 Number(6.0)");
assert!(ir.constants.contains(&Constant::Int(3)), "常量池应含加法折值 Int(3)");
}
/// 面 15:字符串域链折叠命中矩阵——全操作数为可折叠字面量且至少一个字符串
/// 字面量时,Addition 链折单字符串常量:ADD/CONCAT_N 计数为 0、池含折值键。
#[test]
fn string_chain_fold_hit() {
// 纯字符串链:池含 "ab"。
let ir = emit_ir("var s = \"a\" + \"b\";");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0, "全字面量字符串链应折叠,ADD 计数为 0");
assert_eq!(count_op(&ir.insts, OpCode::CONCAT_N), 0, "全字面量字符串链应折叠,CONCAT_N 计数为 0");
assert!(pool_strings(&ir).contains("ab"), "池应含折值 ab");
// 字符串加数字加布尔加 null:池含 "a1truenull"。
let ir = emit_ir("var s = \"a\" + 1 + true + null;");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0);
assert_eq!(count_op(&ir.insts, OpCode::CONCAT_N), 0);
assert!(pool_strings(&ir).contains("a1truenull"), "池应含折值 a1truenull");
// 字符串加 BigInt:池含 "x10"。
let ir = emit_ir("var s = \"x\" + 10n;");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0);
assert!(pool_strings(&ir).contains("x10"), "池应含折值 x10");
// 字符串加非整数字面量(Number 臂走 js_number_to_string):池含 "a1.5"。
let ir = emit_ir("var s = \"a\" + 1.5;");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0);
assert!(pool_strings(&ir).contains("a1.5"), "池应含折值 a1.5");
// 字符串加 -0(is_int_literal 为真走 Int 臂):池含 "a0"。
let ir = emit_ir("var s = \"a\" + -0;");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0);
assert!(pool_strings(&ir).contains("a0"), "池应含折值 a0");
// 孤立 surrogate 字符串段:池键为 FFFD+hex4 marker 形态加 "a"。
let ir = emit_ir("var s = \"\\ud800\" + \"a\";");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0);
let pool = pool_strings(&ir);
assert!(pool.contains("\u{FFFD}d800a"), "池键应为 FFFD+hex4 marker 形态加 a: {pool:?}");
// 超平面字符段:经 encode_key 良配对形态入池。
let ir = emit_ir("var s = \"\\u{1F600}\" + \"a\";");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 0);
assert!(pool_strings(&ir).contains("\u{1F600}a"), "池应含超平面字符良配对形态加 a");
}
/// 面 16:字符串域链不折矩阵——含标识符操作数或全数字链不落字符串域,
/// 既有 CONCAT_N/ADD 路径计数不变、池不含折值键。
#[test]
fn string_chain_fold_miss() {
// 标识符操作数(两操作数形态):不折,落回左结合 ADD。
let ir = emit_ir("var x = 1; var s = x + \"a\";");
assert_eq!(count_op(&ir.insts, OpCode::ADD), 1, "含标识符的链应落回 ADD");
assert_eq!(count_op(&ir.insts, OpCode::CONCAT_N), 0);
assert!(!pool_strings(&ir).contains("xa"), "含标识符的链不应折出字符串键");
// 全数字链:数值域守卫排除,不折出字符串键,落回 CONCAT_N。
let ir = emit_ir("var s = 1 + 2 + 3;");
assert_eq!(count_op(&ir.insts, OpCode::CONCAT_N), 1, "全数字链应落回 CONCAT_N");
assert!(!pool_strings(&ir).contains("123"), "全数字链不应折出字符串键 123");
// 字符串夹标识符(三操作数形态):不折,落回 CONCAT_N。
let ir = emit_ir("var x = 1; var s = \"a\" + x + \"b\";");
assert_eq!(count_op(&ir.insts, OpCode::CONCAT_N), 1, "含标识符的链应落回 CONCAT_N");
assert!(!pool_strings(&ir).contains("axb"), "含标识符的链不应折出字符串键");
}
/// 面 17:模板字面量折叠命中矩阵——全内插表达式为可折叠字面量且全部 cooked
/// 为 Some 时折单字符串常量:TEMPLATE_STR 计数为 0、池含折值键。
#[test]
fn template_fold_hit() {
let ir = emit_ir("var t = `a${1}b`;");
assert_eq!(count_op(&ir.insts, OpCode::TEMPLATE_STR), 0, "全字面量模板应折叠,TEMPLATE_STR 计数为 0");
assert!(pool_strings(&ir).contains("a1b"), "池应含折值 a1b");
// 内插 BigInt:池含 "x10y"。
let ir = emit_ir("var t = `x${10n}y`;");
assert_eq!(count_op(&ir.insts, OpCode::TEMPLATE_STR), 0);
assert!(pool_strings(&ir).contains("x10y"), "池应含折值 x10y");
// 无内插表达式的单 quasi 模板:折单串。
let ir = emit_ir("var t = `hello`;");
assert_eq!(count_op(&ir.insts, OpCode::TEMPLATE_STR), 0);
assert!(pool_strings(&ir).contains("hello"), "池应含折值 hello");
// 孤立 surrogate quasi:池键为 marker 形态。
let ir = emit_ir("var t = `\\ud800${1}`;");
assert_eq!(count_op(&ir.insts, OpCode::TEMPLATE_STR), 0);
let pool = pool_strings(&ir);
assert!(pool.contains("\u{FFFD}d8001"), "池键应为 FFFD+hex4 marker 形态加 1: {pool:?}");
}
/// 面 18:模板字面量不折矩阵——内插表达式非字面量时落回 TEMPLATE_STR;
/// 带标签模板走 GET_TEMPLATE_OBJECT 加 CALL 的独立路径,不经折叠。
#[test]
fn template_fold_miss() {
// 表达式非字面量:TEMPLATE_STR 计数 1,池不含折值键。
let ir = emit_ir("var x = 1; var t = `a${x}b`;");
assert_eq!(count_op(&ir.insts, OpCode::TEMPLATE_STR), 1, "非字面量内插应落回 TEMPLATE_STR");
assert!(!pool_strings(&ir).contains("axb"), "不折时池不应含折值键");
// 带标签模板:走 GET_TEMPLATE_OBJECT 加 CALL,不折。
let ir = emit_ir("var tag = (s) => s[0]; var t = tag`a${1}`;");
assert_eq!(count_op(&ir.insts, OpCode::TEMPLATE_STR), 0, "带标签模板不发 TEMPLATE_STR");
assert_eq!(count_op(&ir.insts, OpCode::GET_TEMPLATE_OBJECT), 1, "带标签模板应发 GET_TEMPLATE_OBJECT");
assert_eq!(count_op(&ir.insts, OpCode::CALL), 1, "带标签模板应发 CALL");
}
/// 面 19:字符串域折叠发射确定性——折叠路径纯函数无迭代序依赖,链折叠与
/// 模板折叠源串同进程连续编译指纹全等。
#[test]
fn string_fold_emission_stable() {
assert_emission_stable("var s = \"a\" + 1 + true + null;", false);
assert_emission_stable("var t = `a${1}b`;", false);
}
+4 -3
View File
@@ -75,7 +75,8 @@ fn fold_hit_matrix() {
/// 守卫不折矩阵:断言目标操作码计数为 1(落回运行时路径)。
/// 除零守卫须同时覆盖 Int(0) 与 Number(±0.0);负指数守卫须同时覆盖 Int/Number
/// 负值;BigInt 字面量、字符串字面量、标识符均不折。
/// 负值;BigInt 字面量、标识符均不折;字符串域守卫(全数字链无字符串字面量)
/// 落回 CONCAT_N。
#[test]
fn fold_guard_matrix() {
// (源串, 目标操作码)
@@ -90,8 +91,8 @@ fn fold_guard_matrix() {
("1n + 2n", OpCode::ADD),
("1 + 1n", OpCode::ADD),
("1n < 2n", OpCode::LT),
// 字符串域不折。
("\"a\" + \"b\"", OpCode::ADD),
// 字符串域守卫:全数字链无字符串字面量,不落字符串域,落回 CONCAT_N。
("1 + 2 + 3", OpCode::CONCAT_N),
// 标识符不折。
("1 + x", OpCode::ADD),
];