]>
Commit | Line | Data |
---|---|---|
43e0ac24 | 1 | /* Optimized memchr implementation for POWER8. |
2b778ceb | 2 | Copyright (C) 2017-2021 Free Software Foundation, Inc. |
43e0ac24 RS |
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 | |
5a82c748 | 17 | <https://www.gnu.org/licenses/>. */ |
43e0ac24 RS |
18 | |
19 | #include <sysdep.h> | |
20 | ||
21 | /* void *[r3] memchr (const void *s [r3], int c [r4], size_t n [r5]) */ | |
22 | ||
43e0ac24 RS |
23 | #ifndef MEMCHR |
24 | # define MEMCHR __memchr | |
25 | #endif | |
066020c5 | 26 | .machine power8 |
43e0ac24 RS |
27 | ENTRY_TOCLESS (MEMCHR) |
28 | CALL_MCOUNT 3 | |
29 | dcbt 0, r3 | |
30 | clrrdi r8, r3, 3 | |
31 | insrdi r4, r4, 8, 48 | |
32 | ||
33 | /* Calculate the last acceptable address and check for possible | |
34 | addition overflow by using satured math: | |
35 | r7 = r3 + r5 | |
36 | r7 |= -(r7 < x) */ | |
37 | add r7, r3, r5 | |
38 | subfc r6, r3, r7 | |
39 | subfe r9, r9, r9 | |
40 | extsw r6, r9 | |
41 | or r7, r7, r6 | |
42 | ||
43 | insrdi r4, r4, 16, 32 | |
44 | cmpldi r5, 32 | |
45 | li r9, -1 | |
46 | rlwinm r6, r3, 3, 26, 28 /* Calculate padding. */ | |
47 | insrdi r4, r4, 32, 0 | |
48 | mr r10, r7 | |
49 | addi r7, r7, -1 | |
50 | #ifdef __LITTLE_ENDIAN__ | |
51 | sld r9, r9, r6 | |
52 | #else | |
53 | srd r9, r9, r6 | |
54 | #endif | |
55 | ble L(small_range) | |
56 | andi. r11, r3, 63 | |
57 | beq cr0, L(align_qw) | |
58 | clrldi r11, r3, 61 | |
59 | ld r12, 0(r8) /* Load doubleword from memory. */ | |
60 | cmpb r3, r12, r4 /* Check for BYTEs in DWORD1. */ | |
61 | and r3, r3, r9 | |
62 | clrldi r6, r7, 61 /* Byte count - 1 in last dword. */ | |
63 | clrrdi r7, r7, 3 /* Address of last doubleword. */ | |
64 | cmpldi cr7, r3, 0 /* Does r3 indicate we got a hit? */ | |
65 | bne cr7, L(done) | |
66 | addi r8, r8, 8 | |
67 | addi r5, r5, -8 | |
68 | add r5, r5, r11 | |
69 | ||
70 | /* Are we now aligned to a quadword boundary? */ | |
71 | andi. r11, r8, 15 | |
72 | beq cr0, L(align_qw) | |
73 | ||
74 | /* Handle DWORD to make it QW aligned. */ | |
75 | ld r12, 0(r8) | |
76 | cmpb r3, r12, r4 | |
77 | cmpldi cr7, r3, 0 | |
78 | bne cr7, L(done) | |
79 | addi r5, r5, -8 | |
80 | addi r8, r8, 8 | |
81 | /* At this point, r8 is 16B aligned. */ | |
82 | L(align_qw): | |
83 | vspltisb v0, 0 | |
84 | /* Precompute vbpermq constant. */ | |
85 | vspltisb v10, 3 | |
86 | li r0, 0 | |
87 | lvsl v11, r0, r0 | |
88 | vslb v10, v11, v10 | |
066020c5 | 89 | mtvrd v1, r4 |
43e0ac24 RS |
90 | vspltb v1, v1, 7 |
91 | cmpldi r5, 64 | |
92 | ble L(tail64) | |
93 | /* Are we 64-byte aligned? If so, jump to the vectorized loop. | |
94 | Note: aligning to 64-byte will necessarily slow down performance for | |
95 | strings around 64 bytes in length due to the extra comparisons | |
96 | required to check alignment for the vectorized loop. This is a | |
97 | necessary tradeoff we are willing to take in order to speed up the | |
98 | calculation for larger strings. */ | |
99 | andi. r11, r8, 63 | |
100 | beq cr0, L(preloop_64B) | |
101 | /* In order to begin the 64B loop, it needs to be 64 | |
102 | bytes aligned. So read until it is 64B aligned. */ | |
103 | lvx v4, 0, r8 | |
104 | vcmpequb v6, v1, v4 | |
105 | vcmpequb. v11, v0, v6 | |
106 | bnl cr6, L(found_16B) | |
107 | addi r8, r8, 16 | |
108 | addi r5, r5, -16 | |
109 | ||
110 | andi. r11, r8, 63 | |
111 | beq cr0, L(preloop_64B) | |
112 | lvx v4, 0, r8 | |
113 | vcmpequb v6, v1, v4 | |
114 | vcmpequb. v11, v0, v6 | |
115 | bnl cr6, L(found_16B) | |
116 | addi r8, r8, 16 | |
117 | addi r5, r5, -16 | |
118 | ||
119 | andi. r11, r8, 63 | |
120 | beq cr0, L(preloop_64B) | |
121 | lvx v4, 0, r8 | |
122 | vcmpequb v6, v1, v4 | |
123 | vcmpequb. v11, v0, v6 | |
124 | bnl cr6, L(found_16B) | |
125 | addi r8, r8, 16 | |
126 | addi r5, r5, -16 | |
127 | /* At this point it should be 64B aligned. | |
128 | Prepare for the 64B loop. */ | |
129 | L(preloop_64B): | |
130 | cmpldi r5, 64 /* Check if r5 < 64. */ | |
131 | ble L(tail64) | |
132 | sub r6, r10, r8 | |
133 | srdi r9, r6, 6 /* Number of loop iterations. */ | |
134 | mtctr r9 /* Setup the counter. */ | |
135 | li r11, 16 /* Load required offsets. */ | |
136 | li r9, 32 | |
137 | li r7, 48 | |
138 | ||
139 | /* Handle r5 > 64. Loop over the bytes in strides of 64B. */ | |
140 | .align 4 | |
141 | L(loop): | |
142 | lvx v2, 0, r8 /* Load 4 quadwords. */ | |
143 | lvx v3, r8, r11 | |
144 | lvx v4, v8, r9 | |
145 | lvx v5, v8, r7 | |
146 | vcmpequb v6, v1, v2 | |
147 | vcmpequb v7, v1, v3 | |
148 | vcmpequb v8, v1, v4 | |
149 | vcmpequb v9, v1, v5 | |
150 | vor v11, v6, v7 | |
151 | vor v12, v8, v9 | |
152 | vor v11, v11, v12 /* Compare and merge into one VR for speed. */ | |
153 | vcmpequb. v11, v0, v11 | |
154 | bnl cr6, L(found) | |
155 | addi r8, r8, 64 /* Adjust address for the next iteration. */ | |
156 | bdnz L(loop) | |
157 | clrldi r5, r6, 58 | |
158 | ||
159 | /* Handle remainder of 64B loop or r5 > 64. */ | |
160 | .align 4 | |
161 | L(tail64): | |
162 | cmpldi r5, 0 | |
163 | beq L(null) | |
164 | lvx v4, 0, r8 | |
165 | vcmpequb v6, v1, v4 | |
166 | vcmpequb. v11, v0, v6 | |
167 | bnl cr6, L(found_16B) | |
168 | addi r8, r8, 16 | |
169 | cmpldi cr6, r5, 16 | |
170 | ble cr6, L(null) | |
171 | addi r5, r5, -16 | |
172 | ||
173 | lvx v4, 0, r8 | |
174 | vcmpequb v6, v1, v4 | |
175 | vcmpequb. v11, v0, v6 | |
176 | bnl cr6, L(found_16B) | |
177 | addi r8, r8, 16 | |
178 | cmpldi cr6, r5, 16 | |
179 | ble cr6, L(null) | |
180 | addi r5, r5, -16 | |
181 | ||
182 | lvx v4, 0, r8 | |
183 | vcmpequb v6, v1, v4 | |
184 | vcmpequb. v11, v0, v6 | |
185 | bnl cr6, L(found_16B) | |
186 | addi r8, r8, 16 | |
187 | cmpldi cr6, r5, 16 | |
188 | ble cr6, L(null) | |
189 | addi r5, r5, -16 | |
190 | ||
191 | lvx v4, 0, r8 | |
192 | vcmpequb v6, v1, v4 | |
193 | vcmpequb. v11, v0, v6 | |
194 | bnl cr6, L(found_16B) | |
195 | li r3, 0 | |
196 | blr | |
197 | ||
198 | /* Found a match in 64B loop. */ | |
199 | .align 4 | |
200 | L(found): | |
201 | /* Permute the first bit of each byte into bits 48-63. */ | |
066020c5 RFF |
202 | vbpermq v6, v6, v10 |
203 | vbpermq v7, v7, v10 | |
204 | vbpermq v8, v8, v10 | |
205 | vbpermq v9, v9, v10 | |
43e0ac24 RS |
206 | /* Shift each component into its correct position for merging. */ |
207 | #ifdef __LITTLE_ENDIAN__ | |
208 | vsldoi v7, v7, v7, 2 | |
209 | vsldoi v8, v8, v8, 4 | |
210 | vsldoi v9, v9, v9, 6 | |
211 | #else | |
212 | vsldoi v6, v6, v6, 6 | |
213 | vsldoi v7, v7, v7, 4 | |
214 | vsldoi v8, v8, v8, 2 | |
215 | #endif | |
216 | /* Merge the results and move to a GPR. */ | |
217 | vor v11, v6, v7 | |
218 | vor v4, v9, v8 | |
219 | vor v4, v11, v4 | |
066020c5 | 220 | mfvrd r5, v4 |
43e0ac24 RS |
221 | #ifdef __LITTLE_ENDIAN__ |
222 | addi r6, r5, -1 | |
223 | andc r6, r6, r5 | |
224 | popcntd r6, r6 | |
225 | #else | |
226 | cntlzd r6, r5 /* Count leading zeros before the match. */ | |
227 | #endif | |
228 | add r3, r8, r6 /* Compute final length. */ | |
229 | blr | |
230 | ||
231 | /* Found a match in last 16 bytes. */ | |
232 | .align 4 | |
233 | L(found_16B): | |
234 | /* Permute the first bit of each byte into bits 48-63. */ | |
066020c5 | 235 | vbpermq v6, v6, v10 |
43e0ac24 RS |
236 | /* Shift each component into its correct position for merging. */ |
237 | #ifdef __LITTLE_ENDIAN__ | |
066020c5 | 238 | mfvrd r7, v6 |
43e0ac24 RS |
239 | addi r6, r7, -1 |
240 | andc r6, r6, r7 | |
241 | popcntd r6, r6 | |
242 | #else | |
243 | vsldoi v6, v6, v6, 6 | |
066020c5 | 244 | mfvrd r7, v6 |
43e0ac24 RS |
245 | cntlzd r6, r7 /* Count leading zeros before the match. */ |
246 | #endif | |
247 | add r3, r8, r6 /* Compute final length. */ | |
248 | cmpld r6, r5 | |
249 | bltlr | |
250 | li r3, 0 | |
251 | blr | |
252 | ||
253 | .align 4 | |
254 | /* r3 has the output of the cmpb instruction, that is, it contains | |
255 | 0xff in the same position as BYTE in the original | |
256 | doubleword from the string. Use that to calculate the pointer. | |
257 | We need to make sure BYTE is *before* the end of the range. */ | |
258 | L(done): | |
259 | #ifdef __LITTLE_ENDIAN__ | |
260 | addi r0, r3, -1 | |
261 | andc r0, r0, r3 | |
262 | popcntd r0, r0 /* Count trailing zeros. */ | |
263 | #else | |
264 | cntlzd r0, r3 /* Count leading zeros before the match. */ | |
265 | #endif | |
266 | cmpld r8, r7 /* Are we on the last dword? */ | |
267 | srdi r0, r0, 3 /* Convert leading/trailing zeros to bytes. */ | |
268 | add r3, r8, r0 | |
269 | cmpld cr7, r0, r6 /* If on the last dword, check byte offset. */ | |
270 | bnelr | |
271 | blelr cr7 | |
272 | li r3, 0 | |
273 | blr | |
274 | ||
275 | .align 4 | |
276 | L(null): | |
277 | li r3, 0 | |
278 | blr | |
279 | ||
280 | /* Deals with size <= 32. */ | |
281 | .align 4 | |
282 | L(small_range): | |
283 | cmpldi r5, 0 | |
284 | beq L(null) | |
285 | ld r12, 0(r8) /* Load word from memory. */ | |
286 | cmpb r3, r12, r4 /* Check for BYTE in DWORD1. */ | |
287 | and r3, r3, r9 | |
288 | cmpldi cr7, r3, 0 | |
289 | clrldi r6, r7, 61 /* Byte count - 1 in last dword. */ | |
290 | clrrdi r7, r7, 3 /* Address of last doubleword. */ | |
291 | cmpld r8, r7 /* Are we done already? */ | |
292 | bne cr7, L(done) | |
293 | beqlr | |
294 | ||
295 | ldu r12, 8(r8) | |
296 | cmpb r3, r12, r4 | |
297 | cmpldi cr6, r3, 0 | |
298 | cmpld r8, r7 | |
299 | bne cr6, L(done) /* Found something. */ | |
300 | beqlr /* Hit end of string (length). */ | |
301 | ||
302 | ldu r12, 8(r8) | |
303 | cmpb r3, r12, r4 | |
304 | cmpldi cr6, r3, 0 | |
305 | cmpld r8, r7 | |
306 | bne cr6, L(done) | |
307 | beqlr | |
308 | ||
309 | ldu r12, 8(r8) | |
310 | cmpb r3, r12, r4 | |
311 | cmpldi cr6, r3, 0 | |
312 | cmpld r8, r7 | |
313 | bne cr6, L(done) | |
314 | beqlr | |
315 | ||
316 | ldu r12, 8(r8) | |
317 | cmpb r3, r12, r4 | |
318 | cmpldi cr6, r3, 0 | |
319 | bne cr6, L(done) | |
320 | blr | |
321 | ||
322 | END (MEMCHR) | |
323 | weak_alias (__memchr, memchr) | |
324 | libc_hidden_builtin_def (memchr) |