LMMP 4.3.0
Lamina High-Precision Arithmetic Library
载入中...
搜索中...
未找到
test_asm_boundary.c 文件参考
#include "lmmp/lmmpn.h"
#include "mparam.h"
#include "lmmp/numth.h"
#include "cases.h"
#include "support/lmmp_test_utils.h"
#include <string.h>
test_asm_boundary.c 的引用(Include)关系图:

浏览该文件的源代码.

结构体

struct  overlap_buf_t

函数

static void overlap_buf_check_guard (const overlap_buf_t *ob, int w, const char *tag)
static overlap_buf_t overlap_buf_create (mp_size_t n, size_t g)
static mp_ptr overlap_buf_dst_of (const overlap_buf_t *ob, int w)
static void overlap_buf_free (overlap_buf_t *ob)
static void overlap_buf_refill (overlap_buf_t *ob, int w)
static void overlap_buf_set_src (overlap_buf_t *ob)
void test_asm_boundary (void)
 Copyright (C) 2026 HJimmyK (Jericho Knox)
 TEST_CASE ("asm/boundary", add_nc_sub_nc_all_ones)
 Copyright (C) 2026 HJimmyK (Jericho Knox)
 TEST_CASE ("asm/boundary", div_1_by_one_and_power_of_two)
 TEST_CASE ("asm/boundary", inv_and_mulmod_boundary)
 TEST_CASE ("asm/boundary", mul_1_addmul_submul_extremes)
 TEST_CASE ("asm/boundary", shl_shr_c_overlap_family)
 TEST_CASE ("asm/boundary", shl_shr_inplace)
 TEST_CASE ("asm/boundary", shl_shr_overlap_family)

结构体说明

◆ overlap_buf_t

struct overlap_buf_t

在文件 test_asm_boundary.c 第 122 行定义.

overlap_buf_t 的协作图:
成员变量
mp_ptr buf
size_t cap
size_t g
size_t n
mp_ptr numa
mp_ptr sep
mp_ptr snap
mp_ptr src

函数说明

◆ overlap_buf_check_guard()

void overlap_buf_check_guard ( const overlap_buf_t * ob,
int w,
const char * tag )
static

在文件 test_asm_boundary.c 第 160 行定义.

160 {
161 mp_srcptr dst = overlap_buf_dst_of(ob, w);
162 for (size_t i = 0; i < ob->cap; ++i) {
163 if (w != 100 && ob->buf + i >= dst && ob->buf + i < dst + ob->n)
164 continue;
165 if (ob->buf[i] != ob->snap[i]) {
166 TEST_CHECK_MSG(false, tag);
167 return;
168 }
169 }
170}
const mp_limb_t * mp_srcptr
定义 lmmp.h:97
#define n
#define TEST_CHECK_MSG(cond, msg)
定义 test.h:38
static mp_ptr overlap_buf_dst_of(const overlap_buf_t *ob, int w)

引用了 overlap_buf_t::buf, overlap_buf_t::cap, n, overlap_buf_check_guard(), overlap_buf_dst_of(), overlap_buf_t::snap , 以及 TEST_CHECK_MSG.

被这些函数引用 overlap_buf_check_guard(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ overlap_buf_create()

overlap_buf_t overlap_buf_create ( mp_size_t n,
size_t g )
static

在文件 test_asm_boundary.c 第 127 行定义.

127 {
128 overlap_buf_t ob = {.cap = n + 2 * g, .g = g, .n = n};
129 ob.buf = alloc_limbs(ob.cap);
130 ob.snap = alloc_limbs(ob.cap);
131 ob.sep = alloc_limbs(n);
132 ob.src = alloc_limbs(n);
133 ob.numa = ob.buf + g;
134 return ob;
135}

引用了 overlap_buf_t::buf, overlap_buf_t::cap, n, overlap_buf_t::numa, overlap_buf_create(), overlap_buf_t::sep, overlap_buf_t::snap , 以及 overlap_buf_t::src.

被这些函数引用 overlap_buf_create(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ overlap_buf_dst_of()

mp_ptr overlap_buf_dst_of ( const overlap_buf_t * ob,
int w )
static

在文件 test_asm_boundary.c 第 156 行定义.

156 {
157 return w == 100 ? ob->sep : ob->numa + w;
158}

引用了 overlap_buf_t::numa, overlap_buf_dst_of() , 以及 overlap_buf_t::sep.

被这些函数引用 overlap_buf_check_guard(), overlap_buf_dst_of(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ overlap_buf_free()

void overlap_buf_free ( overlap_buf_t * ob)
static

在文件 test_asm_boundary.c 第 137 行定义.

137 {
138 lmmp_free(ob->buf);
139 lmmp_free(ob->snap);
140 lmmp_free(ob->sep);
141 lmmp_free(ob->src);
142}
void lmmp_free(void *ptr)
内存释放函数(调用lmmp_heap_free_fn)

引用了 overlap_buf_t::buf, lmmp_free(), overlap_buf_free(), overlap_buf_t::sep, overlap_buf_t::snap , 以及 overlap_buf_t::src.

被这些函数引用 overlap_buf_free(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ overlap_buf_refill()

void overlap_buf_refill ( overlap_buf_t * ob,
int w )
static

在文件 test_asm_boundary.c 第 149 行定义.

149 {
150 for (size_t i = 0; i < ob->cap; ++i)
151 ob->buf[i] = 0xA5A5DEADBEEF00F0ull ^ (i * 0x9e37u + (u64)w * 7u);
152 lmmp_copy(ob->numa, ob->src, ob->n);
153 lmmp_copy(ob->snap, ob->buf, ob->cap);
154}
#define lmmp_copy(dst, src, n)
定义 lmmp.h:474
mp_limb_t u64
Copyright (C) 2026 HJimmyK (Jericho Knox)

引用了 overlap_buf_t::buf, overlap_buf_t::cap, lmmp_copy, overlap_buf_t::n, overlap_buf_t::numa, overlap_buf_refill(), overlap_buf_t::snap , 以及 overlap_buf_t::src.

被这些函数引用 overlap_buf_refill(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ overlap_buf_set_src()

void overlap_buf_set_src ( overlap_buf_t * ob)
static

在文件 test_asm_boundary.c 第 144 行定义.

144 {
145 u64 seed = 0x9e3779b97f4a7c15ull ^ (ob->n * 31 + ob->g);
146 random_limbs(ob->src, ob->n, &seed);
147}

引用了 overlap_buf_t::g, overlap_buf_t::n, overlap_buf_set_src() , 以及 overlap_buf_t::src.

被这些函数引用 overlap_buf_set_src(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ test_asm_boundary()

void test_asm_boundary ( void )

Copyright (C) 2026 HJimmyK (Jericho Knox)

This file is part of LMMP.

LMMP is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version.

This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License along with this library. If not, see https://www.gnu.org/licenses/.

SPDX-License-Identifier: LGPL-3.0-or-later

在文件 test_asm_boundary.c 第 318 行定义.

318 {
319 TEST_RUN("asm/boundary", add_nc_sub_nc_all_ones);
320 TEST_RUN("asm/boundary", mul_1_addmul_submul_extremes);
321 TEST_RUN("asm/boundary", shl_shr_inplace);
322 TEST_RUN("asm/boundary", shl_shr_overlap_family);
323 TEST_RUN("asm/boundary", shl_shr_c_overlap_family);
324 TEST_RUN("asm/boundary", div_1_by_one_and_power_of_two);
325 TEST_RUN("asm/boundary", inv_and_mulmod_boundary);
326}
#define TEST_RUN(category, name)
定义 test.h:36

引用了 test_asm_boundary() , 以及 TEST_RUN.

被这些函数引用 main() , 以及 test_asm_boundary().

函数调用图:
这是这个函数的调用关系图:

◆ TEST_CASE() [1/7]

TEST_CASE ( "asm/boundary" ,
add_nc_sub_nc_all_ones  )

Copyright (C) 2026 HJimmyK (Jericho Knox)

This file is part of LMMP.

LMMP is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version.

This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License along with this library. If not, see https://www.gnu.org/licenses/.

SPDX-License-Identifier: LGPL-3.0-or-later

在文件 test_asm_boundary.c 第 31 行定义.

31 {
32 const mp_size_t n = 8;
33 mp_ptr a = alloc_limbs(n);
34 mp_ptr b = alloc_limbs(n);
35 mp_ptr d = alloc_limbs(n);
36 for (mp_size_t i = 0; i < n; ++i) {
37 a[i] = UINT64_MAX;
38 b[i] = 0;
39 }
40 b[0] = 1;
41
42 mp_limb_t cy = lmmp_add_nc_(d, a, b, n, 1);
43 TEST_CHECK_EQ(cy, 1u);
44 for (mp_size_t i = 0; i < n; ++i) TEST_CHECK_EQ(d[i], (i == 0) ? 1u : 0u);
45
46 // a - a - borrow = all ones, borrow 1
47 for (mp_size_t i = 0; i < n; ++i) a[i] = UINT64_MAX;
48 mp_limb_t bw = lmmp_sub_nc_(d, a, a, n, 1);
49 TEST_CHECK_EQ(bw, 1u);
50 for (mp_size_t i = 0; i < n; ++i) TEST_CHECK_EQ(d[i], UINT64_MAX);
51
52 lmmp_free(a); lmmp_free(b); lmmp_free(d);
53}
mp_limb_t * mp_ptr
定义 lmmp.h:96
uint64_t mp_size_t
定义 lmmp.h:93
uint64_t mp_limb_t
定义 lmmp.h:92
mp_limb_t lmmp_add_nc_(mp_ptr dst, mp_srcptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t c)
带进位的n位加法 [dst,n] = [numa,n] + [numb,n] + c
定义 add_n.c:58
mp_limb_t lmmp_sub_nc_(mp_ptr dst, mp_srcptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t c)
带借位的n位减法 [dst,n] = [numa,n] - [numb,n] - c
定义 sub_n.c:67
#define TEST_CHECK_EQ(a, b)
定义 test.h:39

引用了 lmmp_add_nc_(), lmmp_free(), lmmp_sub_nc_(), n, TEST_CASE , 以及 TEST_CHECK_EQ.

函数调用图:

◆ TEST_CASE() [2/7]

TEST_CASE ( "asm/boundary" ,
div_1_by_one_and_power_of_two  )

在文件 test_asm_boundary.c 第 258 行定义.

258 {
259 u64 seed = 0xf00dface12345678ull;
260 const mp_size_t n_values[] = {1, 2, 5, 16};
261 for (size_t n_index = 0; n_index < ARRAY_SIZE(n_values); ++n_index) {
262 mp_size_t n = n_values[n_index];
263 mp_ptr a = alloc_limbs(n), q = alloc_limbs(n + 1), t = alloc_limbs(n + 2);
264 for (mp_size_t i = 0; i < n; ++i) a[i] = xorshift64(&seed);
265
266 // x == 1:商等于原数,余数 0
267 memcpy(q, a, n * 8);
268 u64 r = lmmp_div_1_(q, q, n, 1);
269 TEST_CHECK_EQ(r, 0u);
270 TEST_CHECK_EQ(memcmp(q, a, n * 8), 0);
271
272 // x 为 2 的幂:重建 q*x + r == a
273 u64 x = 1ull << ((seed & 63) ? (seed & 63) : 1);
274 memcpy(q, a, n * 8);
275 r = lmmp_div_1_(q, q, n, x);
276 TEST_CHECK(r < x);
277 u64 hi = lmmp_mul_1_(t, q, n, x);
278 t[n] = hi;
279 u64 ovf = add_small_inplace(t, n + 1, r);
280 (void)ovf;
281 TEST_CHECK(mpn_eq(t, n + 1, a, n));
282
284 }
285}
mp_limb_t lmmp_div_1_(mp_ptr dstq, mp_srcptr numa, mp_size_t na, mp_limb_t x)
单精度数除法
定义 div.c:137
mp_limb_t lmmp_mul_1_(mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_limb_t x)
乘以单limb操作 [dst,na] = [numa,na] * x
定义 mul_1.c:27
#define t
#define ARRAY_SIZE(a)
Copyright (C) 2026 HJimmyK (Jericho Knox)
定义 perf.h:40
#define TEST_CHECK(cond)
定义 test.h:37

引用了 ARRAY_SIZE, lmmp_div_1_(), lmmp_free(), lmmp_mul_1_(), n, t, TEST_CASE, TEST_CHECK , 以及 TEST_CHECK_EQ.

函数调用图:

◆ TEST_CASE() [3/7]

TEST_CASE ( "asm/boundary" ,
inv_and_mulmod_boundary  )

在文件 test_asm_boundary.c 第 291 行定义.

291 {
292 u64 seed = 0x0123456789abcdeful;
293 for (int i = 0; i < 200; ++i) {
294 u64 x = xorshift64(&seed) | LIMB_B_2;
295 u64 inv = lmmp_inv_1_(x);
296 mp_limb_t prod[2];
297 u64 hi = lmmp_mul_1_(prod, &x, 1, inv);
298 prod[1] = hi;
299 u64 p_hi = prod[1] + x; // x*(B+inv) 的高半(无溢出)
300 TEST_CHECK_EQ(p_hi, ~(u64)0);
301 TEST_CHECK(prod[0] >= (u64)(0 - x));
302
303 u64 mod = xorshift64(&seed) | 3;
304 u64 a = xorshift64(&seed) % mod;
305 u64 b = xorshift64(&seed) % mod;
306 u64 qq = 0;
307 u64 r = lmmp_mulmod_ulong_(a, b, mod, &qq);
308 mp_limb_t ab[2], q2[3];
309 lmmp_mullh_(a, b, ab);
310 mp_limb_t abw[2] = {ab[0], ab[1]};
311 u64 r2 = lmmp_div_1_(q2, abw, 2, mod);
312 TEST_CHECK_EQ(r2, r);
313 TEST_CHECK_EQ(q2[0], qq);
314 TEST_CHECK_EQ(norm_len(q2 + 1, 1), (mp_size_t)0);
315 }
316}
#define prod
#define lmmp_mullh_
#define LIMB_B_2
mp_limb_t lmmp_inv_1_(mp_limb_t x)
1阶逆元计算 (inv1)
定义 inv.c:124
#define r2
ulong lmmp_mulmod_ulong_(ulong a, ulong b, ulong mod, ulongp q)
计算两个无符号整数的乘积,对mod取模,商放入 q 中

引用了 LIMB_B_2, lmmp_div_1_(), lmmp_inv_1_(), lmmp_mul_1_(), lmmp_mullh_, lmmp_mulmod_ulong_(), prod, r2, TEST_CASE, TEST_CHECK , 以及 TEST_CHECK_EQ.

函数调用图:

◆ TEST_CASE() [4/7]

TEST_CASE ( "asm/boundary" ,
mul_1_addmul_submul_extremes  )

在文件 test_asm_boundary.c 第 59 行定义.

59 {
60 const mp_size_t n = 6;
61 mp_ptr a = alloc_limbs(n), d = alloc_limbs(n + 1), t = alloc_limbs(n + 2),
62 acc = alloc_limbs(n + 1), acc2 = alloc_limbs(n + 1);
63 for (mp_size_t i = 0; i < n; ++i) a[i] = UINT64_MAX;
64 u64 x = UINT64_MAX;
65
66 mp_limb_t cy = lmmp_mul_1_(d, a, n, x);
67 TEST_CHECK_EQ(d[0], 1ull);
68 for (mp_size_t i = 1; i < n; ++i) TEST_CHECK_EQ(d[i], UINT64_MAX);
69 TEST_CHECK_EQ(cy, UINT64_MAX - 1);
70
71 // addmul:acc(F..F) + prod == prod<<.. 组合路径
72 for (mp_size_t i = 0; i < n; ++i) acc[i] = UINT64_MAX;
73 acc[n] = 0;
74 u64 c1 = lmmp_add_n_(t, acc, d, n) + cy; // 进位=加进位+乘进位
75 memcpy(acc2, acc, n * 8);
76 acc2[n] = 0;
77 u64 c2 = lmmp_addmul_1_(acc2, a, n, x);
78 TEST_CHECK_EQ(memcmp(acc2, t, n * 8), 0);
79 TEST_CHECK_EQ(c2, c1);
80
81 // submul 组合路径:borrow = w_sub + hi
82 u64 w = lmmp_sub_n_(t, acc, d, n);
83 memcpy(acc2, acc, n * 8);
84 acc2[n] = 0;
85 u64 c3 = lmmp_submul_1_(acc2, a, n, x);
86 TEST_CHECK_EQ(memcmp(acc2, t, n * 8), 0);
87 TEST_CHECK_EQ(c3, w + cy);
88
89 lmmp_free(a); lmmp_free(d); lmmp_free(t); lmmp_free(acc); lmmp_free(acc2);
90}
mp_limb_t lmmp_addmul_1_(mp_ptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t b)
乘以单limb并累加操作 [numa,n] += [numb,n] * b
定义 mul_1.c:76
mp_limb_t lmmp_submul_1_(mp_ptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t b)
乘以单limb并累减操作 [numa,n] -= [numb,n] * b
mp_limb_t lmmp_sub_n_(mp_ptr dst, mp_srcptr numa, mp_srcptr numb, mp_size_t n)
无借位的n位减法 [dst,n] = [numa,n] - [numb,n]
定义 sub_n.c:84
mp_limb_t lmmp_add_n_(mp_ptr dst, mp_srcptr numa, mp_srcptr numb, mp_size_t n)
无进位的n位加法 [dst,n] = [numa,n] + [numb,n]
定义 add_n.c:76
#define c1

引用了 c1, lmmp_add_n_(), lmmp_addmul_1_(), lmmp_free(), lmmp_mul_1_(), lmmp_sub_n_(), lmmp_submul_1_(), n, t, TEST_CASE , 以及 TEST_CHECK_EQ.

函数调用图:

◆ TEST_CASE() [5/7]

TEST_CASE ( "asm/boundary" ,
shl_shr_c_overlap_family  )

在文件 test_asm_boundary.c 第 211 行定义.

211 {
212 // _c 变体的错位重叠族(c 满足契约:shl_c 的 c 高 64-k 位为 0,
213 // shr_c 的 c 低 64-k 位为 0);期望 = 分离调用 + 手工并位。
214 u64 seed = 0xc0ffee1234567890ul;
215 for (mp_size_t n = 2; n <= 12; ++n) {
218
219 const size_t k_values[] = {1, 5, 32, 63};
220 for (size_t k_index = 0; k_index < ARRAY_SIZE(k_values); ++k_index) {
221 size_t k = k_values[k_index];
222 mp_limb_t cshl = xorshift64(&seed) & (((mp_limb_t)1 << k) - 1);
223 mp_limb_t cshr = (xorshift64(&seed) >> (64 - k)) << (64 - k);
224 mp_limb_t ret_shl = ob.src[n - 1] >> (64 - k);
225 mp_limb_t ret_shr = ob.src[0] << (64 - k);
226
227 mp_ptr eshl = alloc_limbs(n), eshr = alloc_limbs(n);
228 lmmp_shl_(eshl, ob.src, n, k);
229 eshl[0] |= cshl;
230 lmmp_shr_(eshr, ob.src, n, k);
231 eshr[n - 1] |= cshr;
232
233 const int ws_shl[] = {100, 0}; // shl_c 纯 eqsep,同 shl_
234 for (size_t w_index = 0; w_index < ARRAY_SIZE(ws_shl); ++w_index) {
235 int w = ws_shl[w_index];
236 overlap_buf_refill(&ob, w);
237 mp_limb_t r = lmmp_shl_c_(overlap_buf_dst_of(&ob, w), ob.numa, n, k, cshl);
238 TEST_CHECK_EQ(r, ret_shl);
239 TEST_CHECK_EQ(memcmp(overlap_buf_dst_of(&ob, w), eshl, n * 8), 0);
240 overlap_buf_check_guard(&ob, w, "shl_c overlap family guard");
241 }
242 const int ws_shr[] = {100, 0, -1, -2, -3};
243 for (size_t w_index = 0; w_index < ARRAY_SIZE(ws_shr); ++w_index) {
244 int w = ws_shr[w_index];
245 overlap_buf_refill(&ob, w);
246 mp_limb_t r = lmmp_shr_c_(overlap_buf_dst_of(&ob, w), ob.numa, n, k, cshr);
247 TEST_CHECK_EQ(r, ret_shr);
248 TEST_CHECK_EQ(memcmp(overlap_buf_dst_of(&ob, w), eshr, n * 8), 0);
249 overlap_buf_check_guard(&ob, w, "shr_c overlap family guard");
250 }
251 lmmp_free(eshl);
252 lmmp_free(eshr);
253 }
254 overlap_buf_free(&ob);
255 }
256}
#define k
mp_limb_t lmmp_shr_c_(mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_size_t shr, mp_limb_t c)
带进位的右移操作 [dst,na] = [numa,na]>>shr,dst的高shr位填充c的高shr位
定义 shr.c:39
mp_limb_t lmmp_shr_(mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_size_t shr)
右移操作 [dst,na] = [numa,na]>>shr,dst的高shr位填充0
定义 shr.c:26
mp_limb_t lmmp_shl_c_(mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_size_t shl, mp_limb_t c)
带进位的左移操作 [dst,na] = [numa,na]<<shl,dst的低shl位填充c的低shl位
定义 shl.c:39
mp_limb_t lmmp_shl_(mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_size_t shl)
左移操作 [dst,na] = [numa,na]<<shl,dst的低shl位填充0
定义 shl.c:26
static void overlap_buf_check_guard(const overlap_buf_t *ob, int w, const char *tag)
static void overlap_buf_refill(overlap_buf_t *ob, int w)
static void overlap_buf_set_src(overlap_buf_t *ob)
static void overlap_buf_free(overlap_buf_t *ob)
static overlap_buf_t overlap_buf_create(mp_size_t n, size_t g)

引用了 ARRAY_SIZE, k, lmmp_free(), lmmp_shl_(), lmmp_shl_c_(), lmmp_shr_(), lmmp_shr_c_(), n, overlap_buf_t::numa, overlap_buf_check_guard(), overlap_buf_create(), overlap_buf_dst_of(), overlap_buf_free(), overlap_buf_refill(), overlap_buf_set_src(), overlap_buf_t::src, TEST_CASE , 以及 TEST_CHECK_EQ.

函数调用图:

◆ TEST_CASE() [6/7]

TEST_CASE ( "asm/boundary" ,
shl_shr_inplace  )

在文件 test_asm_boundary.c 第 92 行定义.

92 {
93 u64 seed = 0x1234567890abcdeful;
94 const mp_size_t n = 16;
95 mp_ptr a = alloc_limbs(n), orig = alloc_limbs(n), ref = alloc_limbs(n);
96 for (mp_size_t i = 0; i < n; ++i) a[i] = xorshift64(&seed);
97 memcpy(orig, a, n * 8);
98
99 // 原地调用与分离缓冲调用同结果(库内交叉;每次迭代重置源)
100 const size_t s_values[] = {1, 7, 31, 63};
101 for (size_t s_index = 0; s_index < ARRAY_SIZE(s_values); ++s_index) {
102 size_t s = s_values[s_index];
103 memcpy(a, orig, n * 8);
104 mp_limb_t re = lmmp_shl_(ref, orig, n, s);
105 mp_limb_t r1 = lmmp_shl_(a, a, n, s);
106 TEST_CHECK_EQ(r1, re);
107 TEST_CHECK_EQ(memcmp(a, ref, n * 8), 0);
108
109 memcpy(a, orig, n * 8);
110 re = lmmp_shr_(ref, orig, n, s);
111 r1 = lmmp_shr_(a, a, n, s);
112 TEST_CHECK_EQ(r1, re);
113 TEST_CHECK_EQ(memcmp(a, ref, n * 8), 0);
114 }
115
116 lmmp_free(a); lmmp_free(orig); lmmp_free(ref);
117}
#define r1
#define s

引用了 ARRAY_SIZE, lmmp_free(), lmmp_shl_(), lmmp_shr_(), n, r1, s, TEST_CASE , 以及 TEST_CHECK_EQ.

函数调用图:

◆ TEST_CASE() [7/7]

TEST_CASE ( "asm/boundary" ,
shl_shr_overlap_family  )

在文件 test_asm_boundary.c 第 172 行定义.

172 {
173 // shl 要求 eqsep,shr 允许 dst <= numa。
174 // 期望值取分离缓冲调用;返回值由输入 limbs 直接推出。
175 for (mp_size_t n = 1; n <= 24; ++n) {
178
179 for (size_t k = 0; k < 64; ++k) {
180 mp_ptr eshl = alloc_limbs(n), eshr = alloc_limbs(n);
181 mp_limb_t ret_shl = k ? (ob.src[n - 1] >> (64 - k)) : 0;
182 mp_limb_t ret_shr = k ? (ob.src[0] << (64 - k)) : 0;
183 lmmp_shl_(eshl, ob.src, n, k);
184 lmmp_shr_(eshr, ob.src, n, k);
185
186 const int ws_shl[] = {100, 0}; // shl 纯 eqsep:部分重叠族契约外,不做断言
187 for (size_t w_index = 0; w_index < ARRAY_SIZE(ws_shl); ++w_index) {
188 int w = ws_shl[w_index];
189 overlap_buf_refill(&ob, w);
190 mp_limb_t r = lmmp_shl_(overlap_buf_dst_of(&ob, w), ob.numa, n, k);
191 TEST_CHECK_EQ(r, ret_shl);
192 TEST_CHECK_EQ(memcmp(overlap_buf_dst_of(&ob, w), eshl, n * 8), 0);
193 overlap_buf_check_guard(&ob, w, "shl overlap family guard");
194 }
195 const int ws_shr[] = {100, 0, -1, -2, -3};
196 for (size_t w_index = 0; w_index < ARRAY_SIZE(ws_shr); ++w_index) {
197 int w = ws_shr[w_index];
198 overlap_buf_refill(&ob, w);
199 mp_limb_t r = lmmp_shr_(overlap_buf_dst_of(&ob, w), ob.numa, n, k);
200 TEST_CHECK_EQ(r, ret_shr);
201 TEST_CHECK_EQ(memcmp(overlap_buf_dst_of(&ob, w), eshr, n * 8), 0);
202 overlap_buf_check_guard(&ob, w, "shr overlap family guard");
203 }
204 lmmp_free(eshl);
205 lmmp_free(eshr);
206 }
207 overlap_buf_free(&ob);
208 }
209}

引用了 ARRAY_SIZE, k, lmmp_free(), lmmp_shl_(), lmmp_shr_(), n, overlap_buf_t::numa, overlap_buf_check_guard(), overlap_buf_create(), overlap_buf_dst_of(), overlap_buf_free(), overlap_buf_refill(), overlap_buf_set_src(), overlap_buf_t::src, TEST_CASE , 以及 TEST_CHECK_EQ.

函数调用图: