1 | /* Vector optimized 32/64 bit S/390 version of wcsrchr. |
2 | Copyright (C) 2015-2022 Free Software Foundation, Inc. |
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 |
17 | <https://www.gnu.org/licenses/>. */ |
18 | |
19 | #include <ifunc-wcsrchr.h> |
20 | #if HAVE_WCSRCHR_Z13 |
21 | |
22 | # include "sysdep.h" |
23 | # include "asm-syntax.h" |
24 | |
25 | .text |
26 | |
27 | /* wchar_t *wcsrchr (const wchar_t *s, wchar_t c) |
28 | Locate the last character c in string. |
29 | |
30 | Register usage: |
31 | -r0=loaded bytes in first part of s. |
32 | -r1=pointer to last occurence of c or NULL if not found. |
33 | -r2=s |
34 | -r3=c |
35 | -r4=tmp |
36 | -r5=current_len |
37 | -v16=part of s |
38 | -v17=index of found element |
39 | -v18=replicated c |
40 | -v19=part of s with last occurence of c. |
41 | -v20=permute pattern |
42 | */ |
43 | ENTRY(WCSRCHR_Z13) |
44 | .machine "z13" |
45 | .machinemode "zarch_nohighgprs" |
46 | |
47 | vlbb %v16,0(%r2),6 /* Load s until next 4k-byte boundary. */ |
48 | lcbb %r0,0(%r2),6 /* Get bytes to 4k-byte boundary or 16. */ |
49 | |
50 | tmll %r2,3 /* Test if s is 4-byte aligned? */ |
51 | jne .Lfallback /* And use common-code variant if not. */ |
52 | |
53 | vlvgf %v18,%r3,0 /* Generate vector which elements are all c. */ |
54 | vrepf %v18,%v18,0 |
55 | |
56 | lghi %r1,-1 /* Currently no c found. */ |
57 | lghi %r5,0 /* current_len = 0. */ |
58 | |
59 | vfeezfs %v17,%v16,%v18 /* Find element equal or zero. */ |
60 | vlgvb %r4,%v17,7 /* Load byte index of c/zero or 16. */ |
61 | clrjl %r4,%r0,.Lfound_first_part /* Found c/zero in loaded bytes. */ |
62 | .Lalign: |
63 | /* Align s to 16 byte. */ |
64 | risbgn %r4,%r2,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */ |
65 | lghi %r5,16 /* current_len = 16. */ |
66 | slr %r5,%r4 /* Compute bytes to 16bytes boundary. */ |
67 | |
68 | .Lloop: |
69 | vl %v16,0(%r5,%r2) /* Load s. */ |
70 | vfeezfs %v17,%v16,%v18 /* Find element equal with zero search. */ |
71 | jno .Lfound /* Found c/zero (cc=0|1|2). */ |
72 | vl %v16,16(%r5,%r2) |
73 | vfeezfs %v17,%v16,%v18 |
74 | jno .Lfound16 |
75 | vl %v16,32(%r5,%r2) |
76 | vfeezfs %v17,%v16,%v18 |
77 | jno .Lfound32 |
78 | vl %v16,48(%r5,%r2) |
79 | vfeezfs %v17,%v16,%v18 |
80 | jno .Lfound48 |
81 | |
82 | aghi %r5,64 |
83 | j .Lloop /* No character and no zero -> loop. */ |
84 | |
85 | .Lfound48: |
86 | la %r5,16(%r5) /* Use la since aghi would clobber cc. */ |
87 | .Lfound32: |
88 | la %r5,16(%r5) |
89 | .Lfound16: |
90 | la %r5,16(%r5) |
91 | .Lfound: |
92 | je .Lzero /* Found zero, but no c before that zero. */ |
93 | /* Save this part of s to check for further matches after reaching |
94 | the end of the complete string. */ |
95 | vlr %v19,%v16 |
96 | lgr %r1,%r5 |
97 | |
98 | jh .Lzero /* Found a zero after the found c. */ |
99 | aghi %r5,16 /* Start search of next part of s. */ |
100 | j .Lloop |
101 | |
102 | .Lfound_first_part: |
103 | /* This code is only executed if the found c/zero is whithin loaded |
104 | bytes. If no c/zero was found (cc==3) the found index = 16, thus |
105 | this code is not called. |
106 | Resulting condition code of vector find element equal: |
107 | cc==0: no c, found zero |
108 | cc==1: c found, no zero |
109 | cc==2: c found, found zero after c |
110 | cc==3: no c, no zero (this case can be ignored). */ |
111 | je .Lzero /* Found zero, but no c before that zero. */ |
112 | |
113 | locgrne %r1,%r5 /* Mark c as found in first part of s. */ |
114 | vlr %v19,%v16 |
115 | |
116 | jl .Lalign /* No zero (e.g. if vr was fully loaded) |
117 | -> Align and loop afterwards. */ |
118 | |
119 | /* Found a zero in vr. If vr was not fully loaded due to block |
120 | boundary, the remaining bytes are filled with zero and we can't |
121 | rely on zero indication of condition code here! */ |
122 | |
123 | vfenezf %v17,%v16,%v16 |
124 | vlgvb %r4,%v17,7 /* Load byte index of zero or 16. */ |
125 | clrjl %r4,%r0,.Lzero /* Zero within loaded bytes -> end. */ |
126 | j .Lalign /* Align and loop afterwards. */ |
127 | |
128 | .Lend_searched_zero: |
129 | vlgvb %r4,%v17,7 /* Load byte index of zero. */ |
130 | algr %r5,%r4 |
131 | la %r2,0(%r5,%r2) /* Return pointer to zero. */ |
132 | br %r14 |
133 | |
134 | .Lzero: |
135 | /* Reached end of string. Check if one c was found before. */ |
136 | clije %r3,0,.Lend_searched_zero /* Found zero and c is zero. */ |
137 | |
138 | cgfi %r1,-1 /* No c found -> return NULL. */ |
139 | locghie %r2,0 |
140 | ber %r14 |
141 | |
142 | larl %r3,.Lpermute_mask /* Load permute mask. */ |
143 | vl %v20,0(%r3) |
144 | |
145 | /* c was found and is part of v19. */ |
146 | vfenezf %v17,%v19,%v19 /* Find zero. */ |
147 | vlgvb %r4,%v17,7 /* Load byte index of zero or 16. */ |
148 | ahi %r4,3 /* Found zero index is first byte, |
149 | thus highest byte index is last byte of |
150 | wchar_t zero. */ |
151 | |
152 | clgfi %r5,0 /* Loaded byte count in v19 is 16, ... */ |
153 | lochine %r0,16 /* ... if v19 is not the first part of s. */ |
154 | ahi %r0,-1 /* Convert byte count to highest index. */ |
155 | |
156 | clr %r0,%r4 |
157 | locrl %r4,%r0 /* r4 = min (zero-index, highest-index). */ |
158 | |
159 | /* Right-shift of v19 to mask bytes after zero. */ |
160 | clije %r4,15,.Lzero_permute /* No shift is needed if highest index |
161 | in vr is 15. */ |
162 | lhi %r0,15 |
163 | slr %r0,%r4 /* Compute byte count for vector shift left. */ |
164 | sll %r0,3 /* Convert to bit count. */ |
165 | vlvgb %v17,%r0,7 |
166 | vsrlb %v19,%v19,%v17 /* Vector shift right by byte by number of bytes |
167 | specified in bits 1-4 of byte 7 in v17. */ |
168 | |
169 | /* Reverse bytes in v19. */ |
170 | .Lzero_permute: |
171 | vperm %v19,%v19,%v19,%v20 /* Permute v19 to reversed order. */ |
172 | |
173 | /* Find c in reversed v19. */ |
174 | vfeef %v19,%v19,%v18 /* Find c. */ |
175 | la %r2,0(%r1,%r2) |
176 | vlgvb %r3,%v19,7 /* Load byte index of c. */ |
177 | |
178 | /* Compute index in real s and return. */ |
179 | slgr %r4,%r3 |
180 | lay %r2,-3(%r4,%r2) /* Return pointer to zero. -3 is needed, |
181 | because the found byte index is reversed in |
182 | vector-register. Thus point to first byte of |
183 | wchar_t. */ |
184 | br %r14 |
185 | .Lpermute_mask: |
186 | .byte 0x0C,0x0D,0x0E,0x0F,0x08,0x09,0x0A,0x0B |
187 | .byte 0x04,0x05,0x06,0x07,0x00,0x01,0x02,0x03 |
188 | .Lfallback: |
189 | jg WCSRCHR_C |
190 | END(WCSRCHR_Z13) |
191 | |
192 | # if ! HAVE_WCSRCHR_IFUNC |
193 | strong_alias (WCSRCHR_Z13, wcsrchr) |
194 | # endif |
195 | #endif |
196 | |