1 /* Optimized strncmp implementation for PowerPC64/POWER9.
2 Copyright (C) 2016 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 <http://www.gnu.org/licenses/>. */
18 #ifdef __LITTLE_ENDIAN__
21 /* Implements the function
23 int [r3] strncmp (const char *s1 [r3], const char *s2 [r4], size_t [r5] n)
25 The implementation uses unaligned doubleword access to avoid specialized
26 code paths depending of data alignment for first 32 bytes and uses
27 vectorised loops after that. */
29 /* TODO: Change this to actual instructions when minimum binutils is upgraded
30 to 2.27. Macros are defined below for these newer instructions in order
31 to maintain compatibility. */
32 # define VCTZLSBB(r,v) .long (0x10010602 | ((r)<<(32-11)) | ((v)<<(32-21)))
34 # define VEXTUBRX(t,a,b) .long (0x1000070d \
39 # define VCMPNEZB(t,a,b) .long (0x10000507 \
44 /* Get 16 bytes for unaligned case.
45 reg1: Vector to hold next 16 bytes.
46 reg2: Address to read from.
47 reg3: Permute control vector. */
48 # define GET16BYTES(reg1, reg2, reg3) \
50 vperm v8, v2, reg1, reg3; \
51 vcmpequb. v8, v0, v8; \
62 vperm reg1, v9, reg1, reg3;
64 /* TODO: change this to .machine power9 when minimum binutils
65 is upgraded to 2.27. */
67 EALIGN (strncmp, 4, 0)
68 /* Check if size is 0. */
73 /* Check if [s1]+32 or [s2]+32 will cross a 4K page boundary using
76 (((size_t) s1) % PAGE_SIZE > (PAGE_SIZE - ITER_SIZE))
78 with PAGE_SIZE being 4096 and ITER_SIZE begin 32. */
80 cmpldi cr7, r8, 4096-32
83 cmpldi cr7, r9, 4096-32
86 /* For short strings up to 32 bytes, load both s1 and s2 using
87 unaligned dwords and compare. */
95 bne cr0, L(different1)
97 /* If the strings compared are equal, but size is less or equal
109 bne cr0, L(different1)
113 /* Update pointers and size. */
124 bne cr0, L(different1)
135 bne cr0, L(different1)
140 /* Update pointers and size. */
145 /* Now it has checked for first 32 bytes, align source1 to doubleword
146 and adjust source2 address. */
152 lvsr v6, 0, r4 /* Compute mask. */
157 /* Both s1 and s2 are unaligned. */
158 GET16BYTES(v5, r4, v6)
159 lvsr v10, 0, r3 /* Compute mask. */
162 GET16BYTES(v4, r3, v10)
167 /* Align s1 to qw and adjust s2 address. */
180 /* There are 2 loops depending on the input alignment.
181 Each loop gets 16 bytes from s1 and s2, checks for null
182 and compares them. Loops until a mismatch or null occurs. */
185 GET16BYTES(v5, r4, v6)
187 bne cr6, L(different)
195 GET16BYTES(v5, r4, v6)
197 bne cr6, L(different)
205 GET16BYTES(v5, r4, v6)
207 bne cr6, L(different)
215 GET16BYTES(v5, r4, v6)
217 bne cr6, L(different)
229 bne cr6, L(different)
239 bne cr6, L(different)
249 bne cr6, L(different)
259 bne cr6, L(different)
266 /* Calculate and return the difference. */
284 /* The code now checks if r8 and r5 are different by issuing a
285 cmpb and shifts the result based on its output:
287 leadzero = (__builtin_ffsl (z1) - 1);
288 leadzero = leadzero > (n-1)*8 ? (n-1)*8 : leadzero;
289 r1 = (r1 >> leadzero) & 0xFFUL;
290 r2 = (r2 >> leadzero) & 0xFFUL;
303 ble cr7, L(different2)
316 /* If unaligned 16 bytes reads across a 4K page boundary, it uses
317 a simple byte a byte comparison until the page alignment for s1
325 bne cr7, L(byte_ne_3)
327 beq cr7, L(byte_ne_0)
343 bne cr7, L(byte_ne_2)
344 beq cr5, L(byte_ne_0)
349 bdnz L(pagecross_loop0)
372 libc_hidden_builtin_def(strncmp)
374 #include <sysdeps/powerpc/powerpc64/power8/strncmp.S>