1 /* Vector optimized 32/64 bit S/390 version of wcsncpy.
2 Copyright (C) 2015 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/>. */
19 #if defined HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc)
22 # include "asm-syntax.h"
26 /* wchar_t *wcsncpy (const wchar_t *dest, const wchar_t *src, size_t n)
27 Copy at most n characters of string src to dest.
30 -r0=dest pointer for return
31 -r1=tmp, zero byte index
41 -v31=register save area for r6, r7
45 .machinemode "zarch_nohighgprs"
47 # if !defined __s390x__
49 # endif /* !defined __s390x__ */
52 ber %r14 /* Nothing to do, if n == 0. */
54 vlbb %v16,0(%r3),6 /* Load s until next 4k-byte boundary. */
56 tmll %r3,3 /* Test if s is 4-byte aligned? */
57 jne .Lfallback /* And use common-code variant if not. */
59 vlvgp %v31,%r6,%r7 /* Save registers. */
60 lgr %r0,%r2 /* Save destination pointer for return. */
62 lcbb %r6,0(%r3),6 /* Get bytes to 4k-byte boundary or 16. */
63 llgfr %r6,%r6 /* Convert 32bit to 64bit. */
65 lghi %r5,0 /* current_len = 0. */
67 /* Check range of maxlen and convert to byte-count. */
69 tmhh %r4,49152 /* Test bit 0 or 1 of n. */
70 lghi %r1,-4 /* Max byte-count is 18446744073709551612. */
72 tmlh %r4,49152 /* Test bit 0 or 1 of n. */
73 llilf %r1,4294967292 /* Max byte-count is 4294967292. */
74 # endif /* !__s390x__ */
75 sllg %r4,%r4,2 /* Convert character-count to byte-count. */
76 locgrne %r4,%r1 /* Use max byte-count, if bit 0/1 was one. */
78 clgrjle %r4,%r6,.Lremaining_v16 /* If n <= loaded-bytes
79 -> process remaining. */
81 /* n > loaded-byte-count. */
82 vfenezf %v17,%v16,%v16 /* Find element not equal with zero search. */
83 vlgvb %r1,%v17,7 /* Load zero index or 16 if not found. */
84 aghi %r1,3 /* Also copy remaining bytes of zero. */
85 clrjl %r1,%r6,.Lfound_v16_store /* Found zero within loaded bytes,
88 /* Align s to 16 byte. */
89 risbgn %r7,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */
90 lghi %r5,15 /* current_len = 15. */
91 slr %r5,%r7 /* Compute highest index to 16byte boundary. */
93 /* Zero not found and n > loaded-byte-count. */
94 vstl %v16,%r5,0(%r2) /* Copy loaded characters - no zero. */
95 ahi %r5,1 /* Start loop at next character. */
97 /* Now we are 16byte aligned, so we can load
98 a full vreg without page fault. */
99 lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */
101 clgrjl %r1,%r4,.Lloop64
103 vl %v16,0(%r5,%r3) /* Load s. */
104 clgijl %r4,17,.Lremaining_v16 /* If n <=16, process remaining
108 slgfi %r7,16 /* border_len = maxlen - 16. */
110 clgrjhe %r5,%r7,.Lremaining_v16 /* If current_len >= border
111 then process remaining bytes. */
112 vfenezfs %v17,%v16,%v16 /* Find element not equal with zero search. */
113 je .Lfound_v16 /* Jump away if zero was found. */
114 vl %v18,16(%r5,%r3) /* Load next part of s. */
115 vst %v16,0(%r5,%r2) /* Store previous part without zero to dst. */
118 clgrjhe %r5,%r7,.Lremaining_v18
119 vfenezfs %v17,%v18,%v18
125 clgrjhe %r5,%r7,.Lremaining_v16
126 vfenezfs %v17,%v16,%v16
135 /* v16 contains the remaining bytes [1...16].
136 Store remaining bytes and append string-termination. */
137 vfenezf %v17,%v16,%v16 /* Find element not equal with zero search. */
138 slgrk %r7,%r4,%r5 /* Remaining bytes = maxlen - current_len. */
139 aghi %r7,-1 /* vstl needs highest index. */
140 la %r2,0(%r5,%r2) /* vstl has no index register. */
141 vlgvb %r1,%v17,7 /* Load zero index or 16 if not found. */
142 aghi %r1,3 /* Also copy remaining bytes of zero. */
143 /* Zero in remaining bytes? -> jump away (zero-index < max-index)
144 Do not jump away if zero-index == max-index,
145 but simply copy zero with vstl below. */
146 clrjl %r1,%r7,.Lfound_v16_store
147 vstl %v16,%r7,0(%r2) /* Store remaining bytes without null
150 /* Restore saved registers. */
153 lgr %r2,%r0 /* Load saved dest-ptr. */
166 /* v16 contains a zero. Store remaining bytes to zero. current_len
167 has not reached border, thus checking for n is not needed! */
168 vlgvb %r1,%v17,7 /* Load byte index of zero. */
169 la %r2,0(%r5,%r2) /* vstl has no support for index-register. */
170 aghi %r1,3 /* Also copy remaining bytes of zero. */
172 vstl %v16,%r1,0(%r2) /* Copy characters including zero. */
173 /* Fill remaining bytes with zero - remaining count always > 0. */
174 algr %r5,%r1 /* Remaining bytes (=%r4) = ... */
175 slgr %r4,%r5 /* = maxlen - (currlen + zero_index + 1). */
176 la %r2,0(%r1,%r2) /* Pointer to zero. start filling beyond. */
177 aghi %r4,-2 /* mvc with exrl needs count - 1.
178 (additional -1, see remaining bytes above) */
179 srlg %r6,%r4,8 /* Split into 256 byte blocks. */
183 mvc 1(256,%r2),0(%r2) /* Fill 256 zeros at once. */
185 brctg %r6,.Lzero_loop256 /* Loop until all blocks are processed. */
192 /* Find zero in 16byte aligned loop. */
194 vl %v16,0(%r5,%r3) /* Load s. */
195 vfenezfs %v17,%v16,%v16 /* Find element not equal with zero search. */
196 je .Lfound_v16 /* Jump away if zero was found. */
197 vl %v18,16(%r5,%r3) /* Load next part of s. */
198 vst %v16,0(%r5,%r2) /* Store previous part without zero to dst. */
199 vfenezfs %v17,%v18,%v18
203 vfenezfs %v17,%v16,%v16
207 vfenezfs %v17,%v18,%v18
212 lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */
214 clgrjl %r1,%r4,.Lloop64
216 vl %v16,0(%r5,%r3) /* Load s. */
223 #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */