1 /* Optimized memset implementation for PowerPC32/POWER7.
2 Copyright (C) 2010-2019 Free Software Foundation, Inc.
3 Contributed by Luis Machado <luisgpm@br.ibm.com>.
4 This file is part of the GNU C Library.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, see
18 <https://www.gnu.org/licenses/>. */
22 /* void * [r3] memset (void *s [r3], int c [r4], size_t n [r5]));
33 mr 10,3 /* Save original argument for later. */
34 mr 7,1 /* Save original r1 for later. */
37 /* Replicate byte to word. */
41 ble cr6,L(small) /* If length <= 8, use short copy code. */
44 ble cr7,L(medium) /* If length < 32, use medium copy code. */
46 /* Save our word twice to create a doubleword that we will later
49 andi. 11,10,7 /* Check alignment of DST. */
59 /* Get DST aligned to 8 bytes. */
68 4: bf 29,L(big_aligned)
80 srwi 9,5,3 /* Number of full doublewords remaining. */
85 /* From this point on, we'll copy 32+ bytes and the value
86 isn't 0 (so we can't use dcbz). */
94 /* Copy 1~3 doublewords so the main loop starts
95 at a multiple of 32 bytes. */
110 blt cr1,L(tail_bytes)
115 1: /* Copy 1 doubleword. */
122 /* First use a 32-bytes loop with stw's to try and avoid the LHS due
123 to the lfd we will do next. Also, ping-pong through r10 and r12
124 to avoid AGEN delays. */
147 bdnz L(big_loop_fast_setup)
152 /* Now that we're probably past the LHS window, use the VSX to
153 speed up the loop. */
154 L(big_loop_fast_setup):
169 bdnz L(big_loop_fast)
176 /* Check for tail bytes. */
177 mr 1,7 /* Restore r1. */
183 /* At this point we have a tail of 0-7 bytes and we know that the
184 destination is doubleword-aligned. */
185 4: /* Copy 4 bytes. */
190 2: /* Copy 2 bytes. */
195 1: /* Copy 1 byte. */
202 /* Special case when value is 0 and we have a long length to deal
203 with. Use dcbz to zero out 128-bytes at a time. Before using
204 dcbz though, we need to get the destination 128-bytes aligned. */
217 /* Get DST aligned to 128 bytes. */
244 1: bf 31,L(huge_aligned)
255 /* Copies 128-bytes at a time. */
262 /* We have a tail of 0~127 bytes to handle. */
263 mr 1,7 /* Restore r1. */
272 /* We have a tail o 1~127 bytes. Copy up to 15 doublewords for
273 speed. We'll handle the resulting tail bytes later. */
307 /* Handle the rest of the tail bytes here. */
328 /* Expanded tree to copy tail bytes without increments. */
363 /* Handle copies of 9~31 bytes. */
366 /* At least 9 bytes to go. */
369 beq L(medium_aligned)
371 /* Force 4-bytes alignment for DST. */
374 1: /* Copy 1 byte. */
379 2: /* Copy 2 bytes. */
380 bf 30,L(medium_aligned)
387 /* At least 6 bytes to go, and DST is word-aligned. */
398 8: /* Copy 8 bytes. */
404 4: /* Copy 4 bytes. */
409 2: /* Copy 2-3 bytes. */
414 1: /* Copy 1 byte. */
420 /* Handles copies of 0~8 bytes. */
431 libc_hidden_builtin_def (memset)