clb_divmod: division and modulo, so digits have something to stand on
The library had add, subtract (with a free compare in its carry), negate,
increment, decrement, multiply, and the bit ops -- but nothing that divides.
Asking for a decimal expansion without it means inlining a division per digit,
which is where the 11k character estimate came from.
clb_divmod N D Q W V T Ci X BUF leaves Q = N div D and N itself as the
remainder, eating D. Plain repeated subtraction:
Q = 0
while N >= D: N -= D ; Q += 1
Two things it took to get right:
* clb_sub_nc eats both operands, and the subtraction cannot alias (measured:
clb_sub_nc A B A Ci with A=15 B=10 gives 0), so each round needs copies of
N and D. Two of the three copies are pure overhead, but they are in the
loop body, which appears once.
* The first round has to be able to do nothing. Setting T = N and Q = -1
before the loop makes the first commit a no-op and lets the phantom count
cancel out, so the probe can live at the *bottom* of the loop and appear
only once. Without that the probe is duplicated and the tape grows by
about 1.3k.
5839 characters with a minimal driver (5845 with the two outputs), 9 cells,
all 76 sampled pairs correct in r8forge_byte_op_test. Runtime is O(N/D): the
worst case 2047/1 crosses the 5M step budget and hangs, so the test keeps
N/D <= 600 and SPEC says so. Division by zero is undefined, D >= 1 is the
caller's job.
Next: clb_digits on top of this, which needs a shared loop plus per-cell
conditional writes (computed addressing does not exist here) rather than four
copies of divmod.
This commit is contained in:
1 parent
b89fcb0588
commit
0b96b76104
5 files changed
+159
-3
No files matched your search
@@ -80,7 +80,7 @@ r8forge_synth_test examples/full_add.r8asm # all eight combinations ok
|
||||
r8forge verify_clb_bit_add_nc.r8asm # 4 cell 加法切片(进位进->出)
|
||||
r8forge verify_clb_bit_inc_nc.r8asm # 3 cell 自增切片,进位同样进->出
|
||||
r8forge_derive verify_loop.r8asm -1 0=5 # 动态循环:跑 5 次,out 应该是 5 0
|
||||
r8forge_byte_op_test # 字节级库:18 个操作穷举/抽样,1.5 秒
|
||||
r8forge_byte_op_test # 字节级库:19 个操作穷举/抽样,1.5 秒
|
||||
r8forge_macro_docs_test # asm/ 里每个宏的注释格式
|
||||
```
|
||||
|
||||
@@ -212,6 +212,7 @@ inline constexpr uint32_t kDefaultCellMask = (1u << kDefaultCellSize) - 1u;
|
||||
| **`clb_add_ip`(a+=b,原地,4 cell)** | `drv/add_ip_byte.r8asm` | **726 字符** | **2048 抽样全过**(O(b) 运行) |
|
||||
| **`clb_mul`(mc×mul → 两 cell 积,10 cell)** | `drv/mul_byte.r8asm` | **3977 字符** | **65536/65536**(O(位宽) 运行) |
|
||||
| **`clb_ones`(数量 → 从末位起 N 个 1,4 cell)** | `drv/ones.r8asm` | **742 字符**(带 output)/ **828**(最小驱动) | **12 个计数全对**(N=0..11) |
|
||||
| **`clb_divmod`(N div D / N mod D,9 cell)** | `drv/divmod.r8asm` | **5839 字符**(最小驱动)/ 5845(带 output) | **76 组抽样全对**(O(N/D) 运行,`2047/1` 会撞步数上限) |
|
||||
| `asm/fibonacci.r8asm` | — | 11959 字符 | 跑出 13 项 `1 1 2 3 5 8 13 21 34 55 89 144 233` |
|
||||
|
||||
`raw_*` 保留操作数(A、B 原样还回来),`clb_*_nc` 把 A、B 吃干净 —— 所以后者能省掉
|
||||
@@ -478,6 +479,7 @@ SPEC 第 136-150 行的标准宏前缀:
|
||||
| `clb_not_ip` / `clb_inc_ip` / `clb_neg_ip` / `clb_dec_ip` / `clb_add_ip` | 同上再加 `_ip`(in place):结果回到原来的 cell,草稿 cell 由调用者传 |
|
||||
| `clb_nonzero` | 破坏性的纯计算(把 B 抄成"非零"标志),不带 `_ip` 后缀 |
|
||||
| `clb_ones` | 破坏性的纯计算(把"数量"展开成从末位起的一串 1,输入被吃干净);结果写在另一个 cell 里,所以也不带 `_ip` |
|
||||
| `clb_divmod` | 破坏性的纯计算(除/取模,商和余数各落在自己的 cell 里);N、D 都被吃干净,所以不带 `_ip` |
|
||||
| `loop` / `endloop` | 结构性的**宏对**(开/收一个运行时循环),不带前缀;不开 anchor 所以能嵌套,收口后 dp 停在计数器上 |
|
||||
| `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`) |
|
||||
|
||||
@@ -394,6 +394,30 @@ A 的 12 个取值全对(`r8forge_byte_op_test` 的 `test_ones`)。
|
||||
* 反过来先摆一个 `2^n` 再 `clb_dec_ip`,也得到 n 个 1(这条比上面的循环短 128 字符的
|
||||
`clb_dec_ip` 反而更长,因为循环体里要一直 `set_1`,见 `asm/ones.r8asm` 的注释)。
|
||||
|
||||
### 7.6 `clb_divmod`:除法与取模
|
||||
|
||||
`clb_divmod N D Q W V T Ci X BUF`:`Q = N div D`、**`N` 结束时就是余数 R**,D 被吃干净。
|
||||
9 个 cell(3 个数据 + 6 个草稿),纸带 **5839 字符**(`drv/divmod.r8asm` 是 5845),
|
||||
76 组抽样全对(`r8forge_byte_op_test` 的 `test_divmod`)。
|
||||
|
||||
算法是最钝的那招:
|
||||
|
||||
Q = 0
|
||||
while N >= D: N -= D ; Q += 1
|
||||
|
||||
两处踩出来的设计:
|
||||
|
||||
1. **`clb_sub_nc` 会把 N 和 D 都吃掉**,而减法**不能用 `S == A` 的别名** —— 实测
|
||||
`clb_sub_nc A B A Ci`(A=15 B=10 Ci=1)直接给出 `0`。所以每轮都得 copy 出 `W = N`、
|
||||
`V = D` 再减,三个 copy 里有两个是"过路费"。
|
||||
2. **首轮要能什么都不做**:`T` 预置成 `N`、`Q` 预置成 `-1`。于是第一轮"提交 `T -> N`"是
|
||||
空操作,`Q += 1` 正好把 −1 补回 0;探测放在循环体**末尾**,所以零轮(`N < D`)和一轮
|
||||
(`N == D`)都对,而且**探测代码只出现一次**(不这么写要多 1.3k)。
|
||||
|
||||
`O(N/D)` 的代价要记住:`2047 / 1` 会撞 500 万步上限(实测 HANG),每轮约 2.4k 步,
|
||||
所以 **N/D 最多到 1500 左右**;`2047 / 2` 是 1023 轮,正常返回 `1023 1`。
|
||||
除零没有定义,调用者保证 `D >= 1`。
|
||||
|
||||
---
|
||||
|
||||
## 8. 原地 `_ip` 家族(结果回到原来的 cell)
|
||||
@@ -475,13 +499,13 @@ clb_add_ip PH Ci S F ; 把进位加进高字节(Ci 被吃)
|
||||
|
||||
| 测试 | 覆盖 | 断言口径 |
|
||||
|---|---|---|
|
||||
| `r8forge_byte_op_test` | 18 个测试项 = 17 个宏(`clb_neg_nc` 的 Ci=1 / Ci=0 各算一项),见 §7.1、§7.2、§7.5、§8、§9 | **按返回值**:穷举 0..255(`clb_add_ip` 2048 抽样、`clb_mul` 65536 全穷举、`clb_ones` 12 个计数全穷举),任一不符即退出码非 0 |
|
||||
| `r8forge_byte_op_test` | 19 个测试项 = 18 个宏(`clb_neg_nc` 的 Ci=1 / Ci=0 各算一项),见 §7.1、§7.2、§7.5、§7.6、§8、§9 | **按返回值**:穷举 0..255(`clb_add_ip` 2048 抽样、`clb_mul` 65536 全穷举、`clb_ones` 12 个计数全穷举、`clb_divmod` 76 组抽样),任一不符即退出码非 0 |
|
||||
| `r8forge_synth_test examples/full_add.r8asm` | `%synth` 端到端:汇编 + 跑 8 种输入 | `all eight combinations ok` |
|
||||
| `r8forge verify_clb_bit_add_nc.r8asm` | 4 cell 加法切片(进位进→出) | 单点:`0+0+Ci` |
|
||||
| `r8forge verify_clb_bit_inc_nc.r8asm` | 3 cell 自增切片 | 单点 |
|
||||
| `r8forge_derive verify_loop.r8asm -1 0=N` | 动态循环:`N=0..11` 时循环体跑 `N` 次(`A = N`、`C = 0`) | 逐值 `out` |
|
||||
| `r8forge verify_set_0.r8asm` | `set_0` 只清 LSB | `-1 0=255 → 254`、`0=128 → 128` |
|
||||
| `r8forge_macro_docs_test` | `asm/` 里 44 个宏的注释格式 | 全绿或报错行号 |
|
||||
| `r8forge_macro_docs_test` | `asm/` 里 45 个宏的注释格式 | 全绿或报错行号 |
|
||||
| `r8forge_tape_run <tape.rot8>` | 第二个解释器,直接跑纸带 | 与 `r8forge_derive` 对答案 |
|
||||
| `r8forge_derive <src> -1/-2` | 拿汇编源码跑,打印 `out` / 非零 cell | 人工/脚本比对 |
|
||||
|
||||
|
||||
@@ -0,0 +1,77 @@
|
||||
%include "asm/raw_sub_nc.r8asm"
|
||||
%include "asm/add_ip.r8asm"
|
||||
|
||||
; 除法/取模:Q = N div D,R = N mod D。
|
||||
;
|
||||
; 用它自己那句老话:**只有纸带长度是目标**。所以这里不写长除法的移位/恢复,
|
||||
; 就是最钝的那招:
|
||||
;
|
||||
; Q = 0
|
||||
; while N >= D: N -= D ; Q += 1
|
||||
;
|
||||
; 循环体在纸带里只出现一次,长度和 N、D 的大小无关;代价是运行 O(N/D)
|
||||
; (N=2047、D=1 时约 2047 轮,是这台机器上最慢的宏,但还在 500 万步以内)。
|
||||
;
|
||||
; 两个关键点,都是踩出来的:
|
||||
;
|
||||
; 1. **`clb_sub_nc` 会把 N 和 D 都吃掉**,而减法不能用 `S == A` 的别名
|
||||
; (实测 `clb_sub_nc A B A Ci` 直接给出 0)。所以每轮都要 copy 出 W = N、
|
||||
; V = D 再减 —— 三个 copy 里有两个是给它交的"过路费"。
|
||||
;
|
||||
; 2. **首轮要能"什么都不做"**:把 T 预置成 N,Q 预置成 -1。于是第一轮的
|
||||
; "提交 T -> N" 是空操作,"Q += 1" 正好把 -1 补回 0;探测放在循环体末尾,
|
||||
; 所以 `N < D`(零轮)和 `N == D`(一轮)都对,而且探测代码只出现一次。
|
||||
; 没有这个技巧就得在循环外面再写一遍探测,纸带多 1.3k。
|
||||
;
|
||||
; 除零没有定义:调用者保证 D >= 1,这里不检查(D = 0 会一直循环)。
|
||||
|
||||
; Macro clb_divmod
|
||||
; %1: Cell N
|
||||
; %2: Cell D
|
||||
; %3: Cell Q
|
||||
; %4: Cell W
|
||||
; %5: Cell V
|
||||
; %6: Cell T
|
||||
; %7: Cell Ci
|
||||
; %8: Cell X
|
||||
; %9: Cell BUF
|
||||
;
|
||||
; N 是被除数,结束时就是余数 R;D 是除数,会被吃干净;Q 是商(宏自己清零后累加)。
|
||||
; W、V、T、Ci、X、BUF 都是草稿 cell,进来时要求是干净的。
|
||||
; 前提 D >= 1(除零没有定义,这里不检查)。
|
||||
%macro clb_divmod 9
|
||||
.anchor before_clb_divmod
|
||||
|
||||
; T = N:第一轮"提交"是空操作
|
||||
.tp %1
|
||||
copy %6 %9
|
||||
|
||||
; Q = -1:第一轮的 +1 把它补回 0
|
||||
erase %3
|
||||
clb_dec_ip %3 %8
|
||||
|
||||
; Ci = 1 只为让底下那个 biz 直接进循环体;
|
||||
; 真正的判断在体尾的 clb_sub_nc 里
|
||||
.tp %7
|
||||
set_1
|
||||
|
||||
biz
|
||||
; ---- 体:提交上一轮减出来的 T,记一个商,再探一次
|
||||
.tp %6
|
||||
copy %1 %9 ; N = T
|
||||
|
||||
clb_inc_ip %3 %8 ; Q += 1
|
||||
|
||||
.tp %1
|
||||
copy %4 %9 ; W = N
|
||||
.tp %2
|
||||
copy %5 %9 ; V = D
|
||||
.tp %7
|
||||
set_1 ; 进位入 = 1,做的是普通减法
|
||||
clb_sub_nc %4 %5 %6 %7 ; T = N - D,Ci = (N >= D)
|
||||
|
||||
.tp %7
|
||||
rnz
|
||||
|
||||
.tp before_clb_divmod
|
||||
%endmacro
|
||||
@@ -0,0 +1,22 @@
|
||||
%include "asm/divmod.r8asm"
|
||||
|
||||
; clb_divmod 的驱动:0 号 cell 放 X、1 号放 D,输出商和余数。
|
||||
;
|
||||
; r8forge_derive drv/divmod.r8asm -2 0=2047 1=10 # 204 7
|
||||
; r8forge_derive drv/divmod.r8asm -2 0=7 1=10 # 0 7(零轮)
|
||||
;
|
||||
; out 依次是商、余数。
|
||||
.var N 0
|
||||
.var D 1
|
||||
.var Q 2
|
||||
.var W 3
|
||||
.var V 4
|
||||
.var T 5
|
||||
.var Ci 6
|
||||
.var X 7
|
||||
.var BUF 8
|
||||
|
||||
clb_divmod N D Q W V T Ci X BUF
|
||||
|
||||
output Q
|
||||
output N
|
||||
@@ -464,6 +464,36 @@ int test_ones(const std::string &file) {
|
||||
return bad;
|
||||
}
|
||||
|
||||
// clb_divmod: Q = N div D, R = N mod D. Repeated subtraction, so the runtime is
|
||||
// O(N/D) and the worst case (D = 1) would cross the step budget: sampled only.
|
||||
int test_divmod(const std::string &file) {
|
||||
const std::string tape = assemble(file);
|
||||
std::cout << "clb_divmod: " << tape.size() << " characters (sampled)\n";
|
||||
const int nums[] = {0, 1, 2, 9, 10, 11, 100, 511, 1024, 2047};
|
||||
const int divs[] = {1, 2, 3, 7, 10, 100, 1000, 2047};
|
||||
int bad = 0, n = 0;
|
||||
for (int a : nums)
|
||||
for (int d : divs) {
|
||||
if (a / d > 600) continue; // keep well inside the 5M step budget
|
||||
machine m;
|
||||
m.cells[0] = uint16_t(a); // N
|
||||
m.cells[1] = uint16_t(d); // D
|
||||
m.run(tape, 5'000'000);
|
||||
n++;
|
||||
const bool ok = m.finished && m.out.size() == 2 && m.out[0] == a / d &&
|
||||
m.out[1] == a % d;
|
||||
if (!ok) {
|
||||
bad++;
|
||||
if (bad <= 3)
|
||||
report("clb_divmod", a, d, m,
|
||||
std::to_string(a / d) + " " + std::to_string(a % d));
|
||||
}
|
||||
}
|
||||
std::cout << " " << (bad ? "FAILED" : "all " + std::to_string(n) + " sampled pairs ok")
|
||||
<< (bad ? " (" + std::to_string(bad) + " wrong)" : "") << "\n";
|
||||
return bad;
|
||||
}
|
||||
|
||||
// clb_add_ip: A = A + B in place, B is eaten. The loop runs B times, so this
|
||||
// one is sampled rather than exhaustive -- 256 addends x 8 accumulators.
|
||||
int test_add_ip(const std::string &file) {
|
||||
@@ -545,6 +575,7 @@ int main(int argc, char **argv) {
|
||||
test_dec_ip(drv + "dec_ip_byte.r8asm") +
|
||||
test_nonzero(drv + "nonzero_byte.r8asm") +
|
||||
test_ones(drv + "ones.r8asm") +
|
||||
test_divmod(drv + "divmod.r8asm") +
|
||||
test_add_ip(drv + "add_ip_byte.r8asm") +
|
||||
test_mul(drv + "mul_byte.r8asm") +
|
||||
test_sub_nc(drv + "raw_sub_nc_byte.r8asm");
|
||||
|
||||
Reference in new issue
Block a user