From adb3c8d71a520655ce45195121794fd9c9e12c11 Mon Sep 17 00:00:00 2001 From: dsh Date: Sat, 7 Nov 2026 08:53:25 +0800 Subject: [PATCH] bit_raw_add_nc is now bit_add_nc, and here is a 3 cell increment slice The rename: the four cells of that slice all carry meaning -- A and B are inputs, the third is a carry that is read on the way in and written on the way out, the fourth is the sum. There is no scratch cell, so SPEC.md's `raw` ("the caller passes the scratch cell") does not apply and the name now says so. Measured before renaming: A=0 B=0 carry-in=1 gives S=1, so the cell is an input, not a spare. The cache survived the rename because the key is model version + params + mode + metric + the resolved table (src/synth.cc), never the block name: "bit_add_nc_synth: cached, 5 blocks". verify_raw_add_nc.r8asm is now verify_add_nc.r8asm. New asm/raw_inc_nc.r8asm, asked as an experiment ("can we even make one"): bit_inc_nc, a 3 cell increment slice -- A, carry in/out, S -- three parameters instead of raw_inc's five. Table 0 4 4 2 0 4 4 2 resolved to a 4 block program that was already in the cache, emitted tape is 54 characters, and all four input combinations check out: A=0 carry=0 -> S=0 carry=0 A=1 carry=0 -> S=1 carry=0 A=0 carry=1 -> S=1 carry=0 A=1 carry=1 -> S=0 carry=1 so S = A ^ Ci and the carry becomes A & Ci, with A eaten -- the same shape as bit_add_nc, and like that one it cannot drive a byte level rotation (the operand is consumed). verify_inc_nc.r8asm drives it. AGENTS.md: the two slices are in the self check list, the verified table and the naming table; the "every macro" doc check now counts 30 macros in 10 files. --- AGENTS.md | 13 ++++++++----- asm/raw_add_nc.r8asm | 10 ++++++---- asm/raw_inc_nc.r8asm | 33 +++++++++++++++++++++++++++++++++ verify_add_nc.r8asm | 12 ++++++++++++ verify_inc_nc.r8asm | 11 +++++++++++ verify_raw_add_nc.r8asm | 11 ----------- 6 files changed, 70 insertions(+), 20 deletions(-) create mode 100644 asm/raw_inc_nc.r8asm create mode 100644 verify_add_nc.r8asm create mode 100644 verify_inc_nc.r8asm delete mode 100644 verify_raw_add_nc.r8asm diff --git a/AGENTS.md b/AGENTS.md index aea4669..5ba5896 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -54,7 +54,8 @@ clang-cl /nologo /std:c++20 /EHsc /O2 /I include src/r8asm.cc src/synth.cc tests ```sh r8forge asm/test.r8asm # 应 exit 0,1666 字符,7/7 命中缓存 r8forge_synth_test examples/full_add.r8asm # all eight combinations ok -r8forge verify_raw_add_nc.r8asm # 4 cell 加法切片驱动 +r8forge verify_add_nc.r8asm # 4 cell 加法切片(进位进->出) +r8forge verify_inc_nc.r8asm # 3 cell 自增切片,进位同样进->出 r8forge_byte_op_test # 字节级库:65536×4 + 512 全过,0.5 秒 r8forge_macro_docs_test # asm/ 里每个宏的注释格式 ``` @@ -137,7 +138,7 @@ void rot8_rnz(rot8_iter *iter) { if ((*(iter->mem_now) & 1) == 1) { ... } } ### 另一位一 cell 的实现在这里被删过一次(别再加回来) -"每个 bit 一个 cell、用 `bit_raw_add_nc` 串成 8 位加法器"这条路能跑通(32 cell, +"每个 bit 一个 cell、用 `bit_add_nc` 串成 8 位加法器"这条路能跑通(32 cell, 1308 字符,65536 组合全对),但它是上面那个错误诊断的产物:既然一个 cell 能放一个字节、 位读得出来,就没必要为每个 bit 花一个 cell。R8Forge 之前也删过一次同样的东西 (交接文档原话:"8 位加法 25 个 cell,没有实用价值")。**要拼字节级运算就用 `raw_*`。** @@ -160,7 +161,8 @@ void rot8_rnz(rot8_iter *iter) { if ((*(iter->mem_now) & 1) == 1) { ... } } | 东西 | 结论 | |---|---| | 库里的 7 个位运算块(and/or/xor/inc/add) | 合成版比手写短:`asm/test.r8asm` 1714 → 1666 字符 | -| `bit_raw_add_nc`(4 cell 忽略进位切片) | flat 代价 **27**,完整全加器切片是 38;8/8 组合正确 | +| `bit_add_nc`(4 cell 忽略进位切片) | flat 代价 **27**,完整全加器切片是 38;8/8 组合正确 | +| `bit_inc_nc`(3 cell 自增切片) | 表 `0 4 4 2 0 4 4 2`(3 参数)缓存里就有,4 blocks;纸带 54 字符,4/4 组合正确 | | `bit_raw_add`(6 cell 全加器切片) | 保留 A、B;字节级 `raw_add` 1647 字符,65536 组合全对 | | `raw_and` / `raw_or` / `raw_xor` / `raw_inc` | 639 / 575 / 463 / 1214 字符,各 65536(inc 512)全对 | | `asm/fibonacci.r8asm` | 8761 字符,跑出 10 项斐波那契,和注释一致 | @@ -201,7 +203,8 @@ SPEC 第 136-150 行的标准宏前缀: |---|---| | `raw_add` / `raw_and` / `raw_or` / `raw_xor` / `raw_inc` | 字节级,且草稿 cell 由调用者传 | | `bit_raw_add`(6 cell) | 位级,且 `%6: buf` 是真正的草稿:内容无关,结束是垃圾 | -| `bit_raw_add_nc`(4 cell) | **四个 cell 都有意义**,没有草稿 cell:第 3 个是"进→出"的进位(`A=0 B=0 进位入=1` 时 `S=1`,实测过)。名字里的 `raw` 是历史遗留,按 SPEC 严格读不该有 —— 但"调用者得多给一个工作 cell"这个宽松读法又说得通,所以先不改名 | +| `bit_add_nc`(4 cell) | **四个 cell 都有意义**,没有草稿 cell:第 3 个是"进→出"的进位(`A=0 B=0 进位入=1` 时 `S=1`,实测过)。所以名字里没有 `raw` | +| `bit_inc_nc`(3 cell) | 同样:A 被吃掉,进位进→出,没有草稿 cell | | `clb_bit_mov` | 位级搬运,并且把源吃掉 | | `bit_copy` / `copy` | 位级/字节级拷贝,缓冲由调用者传(历史命名,少了 `raw`) | | `set_0` / `set_1` / `erase` / `rol` / `step` / `back` / `output` | 基础操作,不带前缀 | @@ -211,7 +214,7 @@ SPEC 第 136-150 行的标准宏前缀: 宏里的注释: ```asm -; Macro bit_raw_add_nc +; Macro bit_add_nc ; %1: CellBit A ; %2: CellBit B ``` diff --git a/asm/raw_add_nc.r8asm b/asm/raw_add_nc.r8asm index 55691fa..58a68f3 100644 --- a/asm/raw_add_nc.r8asm +++ b/asm/raw_add_nc.r8asm @@ -14,22 +14,24 @@ ; 因为 A、B 被吃掉,它配不了轮转式的字节级驱动(那种驱动每轮还要读回操作数); ; 字节级的 `raw_add` 用的是保留 A、B 的 `bit_raw_add`(6 cell)。 -; Macro bit_raw_add_nc +; Macro bit_add_nc ; %1: CellBit A ; %2: CellBit B ; %3: CellBit Ci ; %4: CellBit S ; +; 四个 cell 都有意义(A、B 是输入,进位进->出,S 是输出),没有草稿 cell —— +; 名字里不带 raw 就是为这个(SPEC:`raw` 指"草稿 cell 要调用者传")。 ; 表:S = A^B^Ci,进位 = AB|ACi|BCi;A、B 结束为 0,S 和进位只落在 0/1 上。 -%synth bit_raw_add_nc_synth +%synth bit_add_nc_synth params 4 stat 0 8 8 4 8 4 4 12 0 8 8 4 8 4 4 12 metric flat mode tape %endsynth -%macro bit_raw_add_nc 4 +%macro bit_add_nc 4 .tp %4 set_0 - bit_raw_add_nc_synth %1 %2 %3 %4 + bit_add_nc_synth %1 %2 %3 %4 %endmacro diff --git a/asm/raw_inc_nc.r8asm b/asm/raw_inc_nc.r8asm new file mode 100644 index 0000000..ebcb1e7 --- /dev/null +++ b/asm/raw_inc_nc.r8asm @@ -0,0 +1,33 @@ +%include "asm/lib.r8asm" + +; 忽略独立进位的自增切片:一个操作数 + 一个"进->出"的进位 cell + 和。 +; +; 和 bit_add_nc 的关系:自增就是"加一个 1",但 1 不用 cell 装,所以少一个参数。 +; 切片占 3 个 cell:A、进位、S。A 被吃掉(结束为 0),进位就地在自己的 cell 里更新。 +; +; 两条要知道的: +; 1. %synth 出来的程序只碰每个 cell 的低 K 位,高位不追踪 —— 这是个位级切片, +; 喂进来的 A、进位都必须是 0/1; +; 2. set_0 清的是最低位(`biz`/`rnz` 判 `cell & 1`,见 AGENTS),不会碰 S 的其余位。 +; +; 同样因为 A 被吃掉,它配不了轮转式的字节级驱动;要字节级自增用 raw_inc +; (那是保留 A 的 bit_raw_inc,5 cell)。 + +; Macro bit_inc_nc +; %1: CellBit A +; %2: CellBit Ci +; %3: CellBit S +; +; 表:S = A^Ci,进位 = A&Ci;A 结束为 0。 +%synth bit_inc_nc_synth + params 3 + stat 0 4 4 2 0 4 4 2 + metric flat + mode tape +%endsynth + +%macro bit_inc_nc 3 + .tp %3 + set_0 + bit_inc_nc_synth %1 %2 %3 +%endmacro diff --git a/verify_add_nc.r8asm b/verify_add_nc.r8asm new file mode 100644 index 0000000..361898c --- /dev/null +++ b/verify_add_nc.r8asm @@ -0,0 +1,12 @@ +%include "asm/raw_add_nc.r8asm" + +; bit_add_nc 的单点验证:A、B、进位各喂一位,打出和与进位。 +; 进位是"进->出"同一个 cell,所以 0=0 1=0 2=1 就是 0+0+1。 +.var A 0 +.var B 1 +.var Ci 2 +.var S 3 + +bit_add_nc A B Ci S +output S +output Ci diff --git a/verify_inc_nc.r8asm b/verify_inc_nc.r8asm new file mode 100644 index 0000000..e02c895 --- /dev/null +++ b/verify_inc_nc.r8asm @@ -0,0 +1,11 @@ +%include "asm/raw_inc_nc.r8asm" + +; bit_inc_nc 的单点验证:A、进位各喂一位,打出和与进位输出。 +; out 依次是 S、进位(进->出同一个 cell)。 +.var A 0 +.var Ci 1 +.var S 2 + +bit_inc_nc A Ci S +output S +output Ci diff --git a/verify_raw_add_nc.r8asm b/verify_raw_add_nc.r8asm deleted file mode 100644 index c4011e7..0000000 --- a/verify_raw_add_nc.r8asm +++ /dev/null @@ -1,11 +0,0 @@ -%include "asm/raw_add_nc.r8asm" - -; bit_raw_add_nc 的单点验证:A、B、进位各喂一位,打出和与进位。 -.var A 0 -.var B 1 -.var Ci 2 -.var S 3 - -bit_raw_add_nc A B Ci S -output S -output Ci