1 | /* Copyright (C) 2011-2021 Free Software Foundation, Inc. |
2 | Contributed by Intel Corporation. |
3 | This file is part of the GNU C Library. |
4 | |
5 | The GNU C Library is free software; you can redistribute it and/or |
6 | modify it under the terms of the GNU Lesser General Public |
7 | License as published by the Free Software Foundation; either |
8 | version 2.1 of the License, or (at your option) any later version. |
9 | |
10 | The GNU C Library is distributed in the hope that it will be useful, |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 | Lesser General Public License for more details. |
14 | |
15 | You should have received a copy of the GNU Lesser General Public |
16 | License along with the GNU C Library; if not, see |
17 | <https://www.gnu.org/licenses/>. */ |
18 | |
19 | #include <sysdep.h> |
20 | |
21 | #ifdef USE_AS_WMEMCHR |
22 | # define MEMCHR wmemchr |
23 | # define PCMPEQ pcmpeqd |
24 | # define CHAR_PER_VEC 4 |
25 | #else |
26 | # define MEMCHR memchr |
27 | # define PCMPEQ pcmpeqb |
28 | # define CHAR_PER_VEC 16 |
29 | #endif |
30 | |
31 | /* fast SSE2 version with using pmaxub and 64 byte loop */ |
32 | |
33 | .text |
34 | ENTRY(MEMCHR) |
35 | movd %esi, %xmm1 |
36 | mov %edi, %ecx |
37 | |
38 | #ifdef __ILP32__ |
39 | /* Clear the upper 32 bits. */ |
40 | movl %edx, %edx |
41 | #endif |
42 | #ifdef USE_AS_WMEMCHR |
43 | test %RDX_LP, %RDX_LP |
44 | jz L(return_null) |
45 | #else |
46 | punpcklbw %xmm1, %xmm1 |
47 | test %RDX_LP, %RDX_LP |
48 | jz L(return_null) |
49 | punpcklbw %xmm1, %xmm1 |
50 | #endif |
51 | |
52 | and $63, %ecx |
53 | pshufd $0, %xmm1, %xmm1 |
54 | |
55 | cmp $48, %ecx |
56 | ja L(crosscache) |
57 | |
58 | movdqu (%rdi), %xmm0 |
59 | PCMPEQ %xmm1, %xmm0 |
60 | pmovmskb %xmm0, %eax |
61 | test %eax, %eax |
62 | |
63 | jnz L(matches_1) |
64 | sub $CHAR_PER_VEC, %rdx |
65 | jbe L(return_null) |
66 | add $16, %rdi |
67 | and $15, %ecx |
68 | and $-16, %rdi |
69 | #ifdef USE_AS_WMEMCHR |
70 | shr $2, %ecx |
71 | #endif |
72 | add %rcx, %rdx |
73 | sub $(CHAR_PER_VEC * 4), %rdx |
74 | jbe L(exit_loop) |
75 | jmp L(loop_prolog) |
76 | |
77 | .p2align 4 |
78 | L(crosscache): |
79 | and $15, %ecx |
80 | and $-16, %rdi |
81 | movdqa (%rdi), %xmm0 |
82 | |
83 | PCMPEQ %xmm1, %xmm0 |
84 | /* Check if there is a match. */ |
85 | pmovmskb %xmm0, %eax |
86 | /* Remove the leading bytes. */ |
87 | sar %cl, %eax |
88 | test %eax, %eax |
89 | je L(unaligned_no_match) |
90 | /* Check which byte is a match. */ |
91 | bsf %eax, %eax |
92 | #ifdef USE_AS_WMEMCHR |
93 | mov %eax, %esi |
94 | shr $2, %esi |
95 | sub %rsi, %rdx |
96 | #else |
97 | sub %rax, %rdx |
98 | #endif |
99 | jbe L(return_null) |
100 | add %rdi, %rax |
101 | add %rcx, %rax |
102 | ret |
103 | |
104 | .p2align 4 |
105 | L(unaligned_no_match): |
106 | /* "rcx" is less than 16. Calculate "rdx + rcx - 16" by using |
107 | "rdx - (16 - rcx)" instead of "(rdx + rcx) - 16" to void |
108 | possible addition overflow. */ |
109 | neg %rcx |
110 | add $16, %rcx |
111 | #ifdef USE_AS_WMEMCHR |
112 | shr $2, %ecx |
113 | #endif |
114 | sub %rcx, %rdx |
115 | jbe L(return_null) |
116 | add $16, %rdi |
117 | sub $(CHAR_PER_VEC * 4), %rdx |
118 | jbe L(exit_loop) |
119 | |
120 | .p2align 4 |
121 | L(loop_prolog): |
122 | movdqa (%rdi), %xmm0 |
123 | PCMPEQ %xmm1, %xmm0 |
124 | pmovmskb %xmm0, %eax |
125 | test %eax, %eax |
126 | jnz L(matches) |
127 | |
128 | movdqa 16(%rdi), %xmm2 |
129 | PCMPEQ %xmm1, %xmm2 |
130 | pmovmskb %xmm2, %eax |
131 | test %eax, %eax |
132 | jnz L(matches16) |
133 | |
134 | movdqa 32(%rdi), %xmm3 |
135 | PCMPEQ %xmm1, %xmm3 |
136 | pmovmskb %xmm3, %eax |
137 | test %eax, %eax |
138 | jnz L(matches32) |
139 | |
140 | movdqa 48(%rdi), %xmm4 |
141 | PCMPEQ %xmm1, %xmm4 |
142 | add $64, %rdi |
143 | pmovmskb %xmm4, %eax |
144 | test %eax, %eax |
145 | jnz L(matches0) |
146 | |
147 | test $0x3f, %rdi |
148 | jz L(align64_loop) |
149 | |
150 | sub $(CHAR_PER_VEC * 4), %rdx |
151 | jbe L(exit_loop) |
152 | |
153 | movdqa (%rdi), %xmm0 |
154 | PCMPEQ %xmm1, %xmm0 |
155 | pmovmskb %xmm0, %eax |
156 | test %eax, %eax |
157 | jnz L(matches) |
158 | |
159 | movdqa 16(%rdi), %xmm2 |
160 | PCMPEQ %xmm1, %xmm2 |
161 | pmovmskb %xmm2, %eax |
162 | test %eax, %eax |
163 | jnz L(matches16) |
164 | |
165 | movdqa 32(%rdi), %xmm3 |
166 | PCMPEQ %xmm1, %xmm3 |
167 | pmovmskb %xmm3, %eax |
168 | test %eax, %eax |
169 | jnz L(matches32) |
170 | |
171 | movdqa 48(%rdi), %xmm3 |
172 | PCMPEQ %xmm1, %xmm3 |
173 | pmovmskb %xmm3, %eax |
174 | |
175 | add $64, %rdi |
176 | test %eax, %eax |
177 | jnz L(matches0) |
178 | |
179 | mov %rdi, %rcx |
180 | and $-64, %rdi |
181 | and $63, %ecx |
182 | #ifdef USE_AS_WMEMCHR |
183 | shr $2, %ecx |
184 | #endif |
185 | add %rcx, %rdx |
186 | |
187 | .p2align 4 |
188 | L(align64_loop): |
189 | sub $(CHAR_PER_VEC * 4), %rdx |
190 | jbe L(exit_loop) |
191 | movdqa (%rdi), %xmm0 |
192 | movdqa 16(%rdi), %xmm2 |
193 | movdqa 32(%rdi), %xmm3 |
194 | movdqa 48(%rdi), %xmm4 |
195 | |
196 | PCMPEQ %xmm1, %xmm0 |
197 | PCMPEQ %xmm1, %xmm2 |
198 | PCMPEQ %xmm1, %xmm3 |
199 | PCMPEQ %xmm1, %xmm4 |
200 | |
201 | pmaxub %xmm0, %xmm3 |
202 | pmaxub %xmm2, %xmm4 |
203 | pmaxub %xmm3, %xmm4 |
204 | pmovmskb %xmm4, %eax |
205 | |
206 | add $64, %rdi |
207 | |
208 | test %eax, %eax |
209 | jz L(align64_loop) |
210 | |
211 | sub $64, %rdi |
212 | |
213 | pmovmskb %xmm0, %eax |
214 | test %eax, %eax |
215 | jnz L(matches) |
216 | |
217 | pmovmskb %xmm2, %eax |
218 | test %eax, %eax |
219 | jnz L(matches16) |
220 | |
221 | movdqa 32(%rdi), %xmm3 |
222 | PCMPEQ %xmm1, %xmm3 |
223 | |
224 | PCMPEQ 48(%rdi), %xmm1 |
225 | pmovmskb %xmm3, %eax |
226 | test %eax, %eax |
227 | jnz L(matches32) |
228 | |
229 | pmovmskb %xmm1, %eax |
230 | bsf %eax, %eax |
231 | lea 48(%rdi, %rax), %rax |
232 | ret |
233 | |
234 | .p2align 4 |
235 | L(exit_loop): |
236 | add $(CHAR_PER_VEC * 2), %edx |
237 | jle L(exit_loop_32) |
238 | |
239 | movdqa (%rdi), %xmm0 |
240 | PCMPEQ %xmm1, %xmm0 |
241 | pmovmskb %xmm0, %eax |
242 | test %eax, %eax |
243 | jnz L(matches) |
244 | |
245 | movdqa 16(%rdi), %xmm2 |
246 | PCMPEQ %xmm1, %xmm2 |
247 | pmovmskb %xmm2, %eax |
248 | test %eax, %eax |
249 | jnz L(matches16) |
250 | |
251 | movdqa 32(%rdi), %xmm3 |
252 | PCMPEQ %xmm1, %xmm3 |
253 | pmovmskb %xmm3, %eax |
254 | test %eax, %eax |
255 | jnz L(matches32_1) |
256 | sub $CHAR_PER_VEC, %edx |
257 | jle L(return_null) |
258 | |
259 | PCMPEQ 48(%rdi), %xmm1 |
260 | pmovmskb %xmm1, %eax |
261 | test %eax, %eax |
262 | jnz L(matches48_1) |
263 | xor %eax, %eax |
264 | ret |
265 | |
266 | .p2align 4 |
267 | L(exit_loop_32): |
268 | add $(CHAR_PER_VEC * 2), %edx |
269 | movdqa (%rdi), %xmm0 |
270 | PCMPEQ %xmm1, %xmm0 |
271 | pmovmskb %xmm0, %eax |
272 | test %eax, %eax |
273 | jnz L(matches_1) |
274 | sub $CHAR_PER_VEC, %edx |
275 | jbe L(return_null) |
276 | |
277 | PCMPEQ 16(%rdi), %xmm1 |
278 | pmovmskb %xmm1, %eax |
279 | test %eax, %eax |
280 | jnz L(matches16_1) |
281 | xor %eax, %eax |
282 | ret |
283 | |
284 | .p2align 4 |
285 | L(matches0): |
286 | bsf %eax, %eax |
287 | lea -16(%rax, %rdi), %rax |
288 | ret |
289 | |
290 | .p2align 4 |
291 | L(matches): |
292 | bsf %eax, %eax |
293 | add %rdi, %rax |
294 | ret |
295 | |
296 | .p2align 4 |
297 | L(matches16): |
298 | bsf %eax, %eax |
299 | lea 16(%rax, %rdi), %rax |
300 | ret |
301 | |
302 | .p2align 4 |
303 | L(matches32): |
304 | bsf %eax, %eax |
305 | lea 32(%rax, %rdi), %rax |
306 | ret |
307 | |
308 | .p2align 4 |
309 | L(matches_1): |
310 | bsf %eax, %eax |
311 | #ifdef USE_AS_WMEMCHR |
312 | mov %eax, %esi |
313 | shr $2, %esi |
314 | sub %rsi, %rdx |
315 | #else |
316 | sub %rax, %rdx |
317 | #endif |
318 | jbe L(return_null) |
319 | add %rdi, %rax |
320 | ret |
321 | |
322 | .p2align 4 |
323 | L(matches16_1): |
324 | bsf %eax, %eax |
325 | #ifdef USE_AS_WMEMCHR |
326 | mov %eax, %esi |
327 | shr $2, %esi |
328 | sub %rsi, %rdx |
329 | #else |
330 | sub %rax, %rdx |
331 | #endif |
332 | jbe L(return_null) |
333 | lea 16(%rdi, %rax), %rax |
334 | ret |
335 | |
336 | .p2align 4 |
337 | L(matches32_1): |
338 | bsf %eax, %eax |
339 | #ifdef USE_AS_WMEMCHR |
340 | mov %eax, %esi |
341 | shr $2, %esi |
342 | sub %rsi, %rdx |
343 | #else |
344 | sub %rax, %rdx |
345 | #endif |
346 | jbe L(return_null) |
347 | lea 32(%rdi, %rax), %rax |
348 | ret |
349 | |
350 | .p2align 4 |
351 | L(matches48_1): |
352 | bsf %eax, %eax |
353 | #ifdef USE_AS_WMEMCHR |
354 | mov %eax, %esi |
355 | shr $2, %esi |
356 | sub %rsi, %rdx |
357 | #else |
358 | sub %rax, %rdx |
359 | #endif |
360 | jbe L(return_null) |
361 | lea 48(%rdi, %rax), %rax |
362 | ret |
363 | |
364 | .p2align 4 |
365 | L(return_null): |
366 | xor %eax, %eax |
367 | ret |
368 | END(MEMCHR) |
369 | |
370 | #ifndef USE_AS_WMEMCHR |
371 | strong_alias (memchr, __memchr) |
372 | libc_hidden_builtin_def(memchr) |
373 | #endif |
374 | |