From efe24d93124a21300191ec2bdae7fa72b334f0a9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=B4=AE=E7=94=9F=EF=BC=88=E5=AD=90=E8=99=9A=EF=BC=89?= <2234839456@qq.com> Date: Sat, 25 Jul 2026 02:52:13 +0800 Subject: [PATCH] =?UTF-8?q?perf(cff):=20rewriteCharstring/replaceDictOffse?= =?UTF-8?q?ts=20=E7=94=A8=20Uint8Array+=E5=86=99=E6=8C=87=E9=92=88?= =?UTF-8?q?=E6=9B=BF=E4=BB=A3=20number[]?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit rewriteCharstring(18.7%) 与 replaceDictOffsets(7.8%) 原 out/chunks: number[] 逐字节 push 有装箱开销 + new Uint8Array(out) 二次遍历 / Array.from+spread 多次拷贝。改预分配 Uint8Array + 写指针 wp（CALLSUBR 截断即 wp=stackStart[]），encodeDictInt 内联到写入处消除临时 number[]。 otf-思源8字 subsetOTF min 0.352→0.322ms(-9%)、p50 0.483→0.421ms(-13%)； otf-白狐5字 min 0.058→0.055ms。输出逐字节完全相同（19188B/26536B A/B 一致）。 Co-Authored-By: Claude Opus 4.8 (1M context) --- backend/font_util/cff-subset.ts | 141 ++++++++++++++++++++++---------- 1 file changed, 96 insertions(+), 45 deletions(-) diff --git a/backend/font_util/cff-subset.ts b/backend/font_util/cff-subset.ts index 64d7b0e..690b3eb 100644 --- a/backend/font_util/cff-subset.ts +++ b/backend/font_util/cff-subset.ts @@ -305,8 +305,18 @@ export function rewriteCharstring( newLocalCount: number, ): Uint8Array { const newLocalBias = subrBias(newLocalCount); - const out: number[] = []; - /** 栈：记录每个 operand 在 out 中的起始位置（便于截断重写）。值为原始解析值。 */ + /** + * 优化：预分配 Uint8Array + 写指针 wp 替代 number[] + push。 + * 原 out: number[] 逐字节 push 有装箱开销，且最终 new Uint8Array(out) 要二次遍历转换。 + * Uint8Array 直接写字节，wp 模拟 length（CALLSUBR 截断即 wp = stackStart[...]）。 + * 容量上界：(end-start) 是原始字节长度；CALLSUBR 重写时新 operand 编码（1~5 字节） + * 可能比原 operand（1~5 字节）长，最坏每个 operand 多 4 字节。operand 数 ≤ end-start， + * 故 (end-start)*2 + 16 是安全上界（远超实际，仅预分配不写入多余字节）。 + */ + const cap = ((end - start) << 1) + 16; + const out = new Uint8Array(cap); + let wp = 0; + /** 栈：记录每个 operand 在输出中的起始 wp（便于截断重写）。值为原始解析值。 */ const stackStart: number[] = []; const stackVal: number[] = []; let stemCount = 0; @@ -314,50 +324,57 @@ export function rewriteCharstring( while (p < end) { const b0 = b[p++]; if (b0 === 255) { - stackStart.push(out.length); + stackStart.push(wp); stackVal.push(NaN); - out.push(255, b[p], b[p + 1], b[p + 2], b[p + 3]); + out[wp] = 255; out[wp + 1] = b[p]; out[wp + 2] = b[p + 1]; out[wp + 3] = b[p + 2]; out[wp + 4] = b[p + 3]; + wp += 5; p += 4; } else if (b0 === 28) { - stackStart.push(out.length); + stackStart.push(wp); stackVal.push(((b[p] << 24) | (b[p + 1] << 16)) >> 16); - out.push(28, b[p], b[p + 1]); + out[wp] = 28; out[wp + 1] = b[p]; out[wp + 2] = b[p + 1]; + wp += 3; p += 2; } else if (b0 === 29) { - stackStart.push(out.length); + stackStart.push(wp); stackVal.push(((b[p] << 24) | (b[p + 1] << 16) | (b[p + 2] << 8) | b[p + 3]) | 0); - out.push(29, b[p], b[p + 1], b[p + 2], b[p + 3]); + out[wp] = 29; out[wp + 1] = b[p]; out[wp + 2] = b[p + 1]; out[wp + 3] = b[p + 2]; out[wp + 4] = b[p + 3]; + wp += 5; p += 4; } else if (b0 >= 32 && b0 <= 246) { - stackStart.push(out.length); + stackStart.push(wp); stackVal.push(b0 - 139); - out.push(b0); + out[wp++] = b0; } else if (b0 >= 247 && b0 <= 250) { - stackStart.push(out.length); + stackStart.push(wp); stackVal.push((b0 - 247) * 256 + b[p] + 108); - out.push(b0, b[p]); + out[wp] = b0; out[wp + 1] = b[p]; + wp += 2; p += 1; } else if (b0 >= 251 && b0 <= 254) { - stackStart.push(out.length); + stackStart.push(wp); stackVal.push(-(b0 - 251) * 256 - b[p] - 108); - out.push(b0, b[p]); + out[wp] = b0; out[wp + 1] = b[p]; + wp += 2; p += 1; } else { /** 操作码 */ if (b0 === 12) { - out.push(12, b[p]); + out[wp] = 12; out[wp + 1] = b[p]; + wp += 2; p += 1; stackStart.length = 0; stackVal.length = 0; } else if (b0 === T2_HSTEM || b0 === T2_VSTEM || b0 === T2_HSTEMHM || b0 === T2_VSTEMHM) { stemCount += stackVal.length >> 1; - out.push(b0); + out[wp++] = b0; stackStart.length = 0; stackVal.length = 0; } else if (b0 === T2_HINTMASK || b0 === T2_CNTRMASK) { - out.push(b0); + out[wp++] = b0; const maskBytes = (stemCount + 7) >>> 3; - for (let i = 0; i < maskBytes; i++) out.push(b[p + i]); + out.set(b.subarray(p, p + maskBytes), wp); + wp += maskBytes; p += maskBytes; stackStart.length = 0; stackVal.length = 0; @@ -367,31 +384,48 @@ export function rewriteCharstring( const newSn = localRemap.get(oldSn); if (newSn === undefined) { /** subr 未保留（理论上引用 charstring 必命中）——保留原 operand 保底 */ - out.push(T2_CALLSUBR); + out[wp++] = T2_CALLSUBR; } else { - /** 截断到栈顶 operand 起始，写入新编号编码 */ - out.length = stackStart[stackStart.length - 1]; - for (const eb of encodeDictInt(newSn - newLocalBias)) out.push(eb); - out.push(T2_CALLSUBR); + /** 截断到栈顶 operand 起始，写入新编号编码（直接写 Uint8Array，不分配临时数组） */ + wp = stackStart[stackStart.length - 1]; + const delta = newSn - newLocalBias; + if (delta >= -107 && delta <= 107) { + out[wp++] = delta + 139; + } else if (delta >= 108 && delta <= 1131) { + const v0 = delta - 108; + out[wp] = 247 + (v0 >> 8); out[wp + 1] = v0 & 0xff; + wp += 2; + } else if (delta >= -1131 && delta <= -108) { + const v0 = -delta - 108; + out[wp] = 251 + (v0 >> 8); out[wp + 1] = v0 & 0xff; + wp += 2; + } else if (delta >= -32768 && delta <= 32767) { + out[wp] = 28; out[wp + 1] = (delta >> 8) & 0xff; out[wp + 2] = delta & 0xff; + wp += 3; + } else { + out[wp] = 29; out[wp + 1] = (delta >>> 24) & 0xff; out[wp + 2] = (delta >> 16) & 0xff; out[wp + 3] = (delta >> 8) & 0xff; out[wp + 4] = delta & 0xff; + wp += 5; + } + out[wp++] = T2_CALLSUBR; } stackStart.length = 0; stackVal.length = 0; } else if (b0 === T2_CALLGSUBR) { /** global subr 不子集化：operand（调用编号）原样保留，bias 不变 */ - out.push(T2_CALLGSUBR); + out[wp++] = T2_CALLGSUBR; stackStart.length = 0; stackVal.length = 0; } else if (b0 === T2_ENDCHAR) { - out.push(b0); + out[wp++] = b0; break; } else { - out.push(b0); + out[wp++] = b0; stackStart.length = 0; stackVal.length = 0; } } } - return new Uint8Array(out); + return out.subarray(0, wp); } /** @@ -1082,8 +1116,15 @@ function encodeCharsetFormat0(cids: number[]): Uint8Array { * @param replacements 操作码键 → 新 offset 值 */ function replaceDictOffsets(dictBytes: Uint8Array, replacements: Map): Uint8Array { - /** 先分段：按操作码切，重组 */ - const chunks: number[][] = []; + /** + * 优化：预分配 Uint8Array + 写指针 wp 替代 chunks: number[][] + Array.from + 二次拼接。 + * 原 chunks 方案每个操作码段分配一个 number[]（含 Array.from 拷贝 + spread 再拷贝）， + * 最后还要两轮遍历拼接。Uint8Array 直接顺序写入，零中间数组。 + * 容量上界：原 DICT 长度 + 每个替换 operand 最大 +4 字节（短编码→长编码）。 + */ + const cap = dictBytes.length + replacements.size * 4 + 16; + const out = new Uint8Array(cap); + let wp = 0; let p = 0; let operandStart = 0; const len = dictBytes.length; @@ -1093,17 +1134,35 @@ function replaceDictOffsets(dictBytes: Uint8Array, replacements: Map= 256) { - chunks.push([...encoded, 12, op & 0xff]); + if (newVal >= -107 && newVal <= 107) { + out[wp++] = newVal + 139; + } else if (newVal >= 108 && newVal <= 1131) { + const v0 = newVal - 108; + out[wp] = 247 + (v0 >> 8); out[wp + 1] = v0 & 0xff; + wp += 2; + } else if (newVal >= -1131 && newVal <= -108) { + const v0 = -newVal - 108; + out[wp] = 251 + (v0 >> 8); out[wp + 1] = v0 & 0xff; + wp += 2; + } else if (newVal >= -32768 && newVal <= 32767) { + out[wp] = 28; out[wp + 1] = (newVal >> 8) & 0xff; out[wp + 2] = newVal & 0xff; + wp += 3; } else { - chunks.push([...encoded, op]); + out[wp] = 29; out[wp + 1] = (newVal >>> 24) & 0xff; out[wp + 2] = (newVal >> 16) & 0xff; out[wp + 3] = (newVal >> 8) & 0xff; out[wp + 4] = newVal & 0xff; + wp += 5; + } + if (op >= 256) { + out[wp] = 12; out[wp + 1] = op & 0xff; + wp += 2; + } else { + out[wp++] = op; } } else { - /** 保留原操作数 + 操作码 */ - chunks.push(Array.from(dictBytes.subarray(operandStart, p))); + /** 保留原操作数 + 操作码（operandStart..p 原样拷贝） */ + out.set(dictBytes.subarray(operandStart, p), wp); + wp += p - operandStart; } operandStart = p; } else if (b0 === 28) { @@ -1122,13 +1181,5 @@ function replaceDictOffsets(dictBytes: Uint8Array, replacements: Map