1 /* Optimized memset implementation for PowerPC64.
2 Copyright (C) 1997-2024 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
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.
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.
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/>. */
21 /* void * [r3] memset (void *s [r3], int c [r4], size_t n [r5]));
24 The memset is done in three sizes: byte (8 bits), word (32 bits),
25 cache line (1024 bits). There is a special case for setting cache lines
26 to 0, to take advantage of the dcbz instruction. */
33 #define rRTN r3 /* Initial value of 1st argument. */
34 #define rMEMP0 r3 /* Original value of 1st arg. */
35 #define rCHR r4 /* Char to set in each byte. */
36 #define rLEN r5 /* Length of region to set. */
37 #define rMEMP r6 /* Address at which we are storing. */
38 #define rALIGN r7 /* Number of bytes we are setting now (when aligning). */
41 #define rNEG64 r8 /* Constant -64 for clearing with dcbz. */
42 #define rCLS r8 /* Cache line size (known to be 128). */
43 #define rCLM r9 /* Cache line size mask to check for cache alignment. */
45 /* Take care of case for size <= 4. */
47 andi. rALIGN, rMEMP0, 3
51 /* Align to word boundary. */
53 insrwi rCHR, rCHR, 8, 16 /* Replicate byte to halfword. */
56 subfic rALIGN, rALIGN, 4
57 add rMEMP, rMEMP, rALIGN
58 sub rLEN, rLEN, rALIGN
65 /* Handle the case of size < 31. */
68 insrwi rCHR, rCHR, 16, 0 /* Replicate halfword to word. */
70 /* Align to 32-byte boundary. */
71 andi. rALIGN, rMEMP, 0x1C
72 subfic rALIGN, rALIGN, 0x20
75 add rMEMP, rMEMP, rALIGN
76 sub rLEN, rLEN, rALIGN
77 cmplwi cr1, rALIGN, 0x10
86 stwu rCHR, -16(rMEMP2)
87 L(a2): bf 29, L(caligned)
90 /* Now aligned to a 32 byte boundary. */
93 clrrwi. rALIGN, rLEN, 5
95 beq cr1, L(zloopstart) /* Special case for clearing memory using dcbz. */
99 beq L(medium) /* We may not actually get to do a full line. */
100 clrlwi. rLEN, rLEN, 27
101 add rMEMP, rMEMP, rALIGN
106 L(c3): dcbtst rNEG64, rMEMP
114 stwu rCHR, -32(rMEMP)
125 stwu rCHR, -32(rMEMP)
127 add rMEMP, rMEMP, rALIGN
131 /* Clear lines of memory in 128-byte chunks. */
133 /* If the remaining length is less the 32 bytes, don't bother getting
134 the cache line size. */
136 li rCLS,128 /* cache line size is 128 */
141 blt cr1,L(handletail32)
155 /* Now we are aligned to the cache line and can use dcbz. */
159 blt cr1,L(handletail32)
165 /* We are here because the cache line size was set and the remainder
166 (rLEN) is less than the actual cache line size.
167 So set up the preconditions for L(nondcbz) and go there. */
169 clrrwi. rALIGN, rLEN, 5
174 /* Memset of 4 bytes or less. */
187 /* Memset of 0-31 bytes. */
192 add rMEMP, rMEMP, rLEN
194 bt- 31, L(medium_31t)
195 bt- 30, L(medium_30t)
197 bt- 29, L(medium_29t)
199 bge- cr1, L(medium_27t)
207 bf- 30, L(medium_30f)
210 bf- 29, L(medium_29f)
213 blt- cr1, L(medium_27f)
218 stwu rCHR, -16(rMEMP)
226 libc_hidden_builtin_def (memset)