]>
Commit | Line | Data |
---|---|---|
04067002 UD |
1 | /* Optimized memset implementation for PowerPC64. |
2 | Copyright (C) 1997,99, 2000,02,03, 2006 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, write to the Free | |
17 | Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA | |
18 | 02110-1301 USA. */ | |
19 | ||
20 | #include <sysdep.h> | |
21 | #include <bp-sym.h> | |
22 | #include <bp-asm.h> | |
23 | ||
24 | /* __ptr_t [r3] memset (__ptr_t s [r3], int c [r4], size_t n [r5])); | |
25 | Returns 's'. | |
26 | ||
27 | The memset is done in three sizes: byte (8 bits), word (32 bits), | |
28 | cache line (1024 bits). There is a special case for setting cache lines | |
29 | to 0, to take advantage of the dcbz instruction. */ | |
30 | ||
31 | EALIGN (BP_SYM (memset), 5, 0) | |
32 | CALL_MCOUNT | |
33 | ||
34 | #define rTMP r0 | |
35 | #define rRTN r3 /* Initial value of 1st argument. */ | |
36 | #define rMEMP0 r3 /* Original value of 1st arg. */ | |
37 | #define rCHR r4 /* Char to set in each byte. */ | |
38 | #define rLEN r5 /* Length of region to set. */ | |
39 | #define rMEMP r6 /* Address at which we are storing. */ | |
40 | #define rALIGN r7 /* Number of bytes we are setting now (when aligning). */ | |
41 | #define rMEMP2 r8 | |
42 | ||
43 | #define rNEG64 r8 /* Constant -64 for clearing with dcbz. */ | |
44 | #define rCLS r8 /* Cache line size (known to be 128). */ | |
45 | #define rCLM r9 /* Cache line size mask to check for cache alignment. */ | |
46 | L(_memset): | |
47 | /* Take care of case for size <= 4. */ | |
48 | cmplwi cr1, rLEN, 4 | |
49 | andi. rALIGN, rMEMP0, 3 | |
50 | mr rMEMP, rMEMP0 | |
51 | ble- cr1, L(small) | |
52 | ||
53 | /* Align to word boundary. */ | |
54 | cmplwi cr5, rLEN, 31 | |
55 | rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ | |
56 | beq+ L(aligned) | |
57 | mtcrf 0x01, rMEMP0 | |
58 | subfic rALIGN, rALIGN, 4 | |
59 | add rMEMP, rMEMP, rALIGN | |
60 | sub rLEN, rLEN, rALIGN | |
61 | bf+ 31, L(g0) | |
62 | stb rCHR, 0(rMEMP0) | |
63 | bt 30, L(aligned) | |
64 | L(g0): | |
65 | sth rCHR, -2(rMEMP) | |
66 | ||
67 | .align 4 | |
68 | /* Handle the case of size < 31. */ | |
69 | L(aligned): | |
70 | mtcrf 0x01, rLEN | |
71 | rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ | |
72 | ble cr5, L(medium) | |
73 | /* Align to 32-byte boundary. */ | |
74 | andi. rALIGN, rMEMP, 0x1C | |
75 | subfic rALIGN, rALIGN, 0x20 | |
76 | beq L(caligned) | |
77 | mtcrf 0x01, rALIGN | |
78 | add rMEMP, rMEMP, rALIGN | |
79 | sub rLEN, rLEN, rALIGN | |
80 | cmplwi cr1, rALIGN, 0x10 | |
81 | mr rMEMP2, rMEMP | |
82 | bf 28, L(a1) | |
83 | stw rCHR, -4(rMEMP2) | |
84 | stwu rCHR, -8(rMEMP2) | |
85 | L(a1): blt cr1, L(a2) | |
86 | stw rCHR, -4(rMEMP2) | |
87 | stw rCHR, -8(rMEMP2) | |
88 | stw rCHR, -12(rMEMP2) | |
89 | stwu rCHR, -16(rMEMP2) | |
90 | L(a2): bf 29, L(caligned) | |
91 | stw rCHR, -4(rMEMP2) | |
92 | ||
93 | .align 4 | |
94 | /* Now aligned to a 32 byte boundary. */ | |
95 | L(caligned): | |
96 | cmplwi cr1, rCHR, 0 | |
97 | clrrwi. rALIGN, rLEN, 5 | |
98 | mtcrf 0x01, rLEN | |
99 | beq cr1, L(zloopstart) /* Special case for clearing memory using dcbz. */ | |
100 | L(nondcbz): | |
101 | srwi rTMP, rALIGN, 5 | |
102 | mtctr rTMP | |
103 | beq L(medium) /* We may not actually get to do a full line. */ | |
104 | clrlwi. rLEN, rLEN, 27 | |
105 | add rMEMP, rMEMP, rALIGN | |
106 | li rNEG64, -0x40 | |
107 | bdz L(cloopdone) | |
108 | ||
109 | .align 4 | |
110 | L(c3): dcbtst rNEG64, rMEMP | |
111 | stw rCHR, -4(rMEMP) | |
112 | stw rCHR, -8(rMEMP) | |
113 | stw rCHR, -12(rMEMP) | |
114 | stw rCHR, -16(rMEMP) | |
115 | stw rCHR, -20(rMEMP) | |
116 | stw rCHR, -24(rMEMP) | |
117 | stw rCHR, -28(rMEMP) | |
118 | stwu rCHR, -32(rMEMP) | |
119 | bdnz L(c3) | |
120 | L(cloopdone): | |
121 | stw rCHR, -4(rMEMP) | |
122 | stw rCHR, -8(rMEMP) | |
123 | stw rCHR, -12(rMEMP) | |
124 | stw rCHR, -16(rMEMP) | |
125 | cmplwi cr1, rLEN, 16 | |
126 | stw rCHR, -20(rMEMP) | |
127 | stw rCHR, -24(rMEMP) | |
128 | stw rCHR, -28(rMEMP) | |
129 | stwu rCHR, -32(rMEMP) | |
130 | beqlr | |
131 | add rMEMP, rMEMP, rALIGN | |
132 | b L(medium_tail2) | |
133 | ||
134 | .align 5 | |
135 | /* Clear lines of memory in 128-byte chunks. */ | |
136 | L(zloopstart): | |
137 | /* If the remaining length is less the 32 bytes, don't bother getting | |
138 | the cache line size. */ | |
139 | beq L(medium) | |
140 | li rCLS,128 /* cache line size is 128 */ | |
141 | dcbt 0,rMEMP | |
142 | L(getCacheAligned): | |
143 | cmplwi cr1,rLEN,32 | |
144 | andi. rTMP,rMEMP,127 | |
145 | blt cr1,L(handletail32) | |
146 | beq L(cacheAligned) | |
147 | addi rMEMP,rMEMP,32 | |
148 | addi rLEN,rLEN,-32 | |
149 | stw rCHR,-32(rMEMP) | |
150 | stw rCHR,-28(rMEMP) | |
151 | stw rCHR,-24(rMEMP) | |
152 | stw rCHR,-20(rMEMP) | |
153 | stw rCHR,-16(rMEMP) | |
154 | stw rCHR,-12(rMEMP) | |
155 | stw rCHR,-8(rMEMP) | |
156 | stw rCHR,-4(rMEMP) | |
157 | b L(getCacheAligned) | |
158 | ||
159 | /* Now we are aligned to the cache line and can use dcbz. */ | |
160 | .align 4 | |
161 | L(cacheAligned): | |
162 | cmplw cr1,rLEN,rCLS | |
163 | blt cr1,L(handletail32) | |
164 | dcbz 0,rMEMP | |
165 | subf rLEN,rCLS,rLEN | |
166 | add rMEMP,rMEMP,rCLS | |
167 | b L(cacheAligned) | |
168 | ||
169 | /* We are here because the cache line size was set and the remainder | |
170 | (rLEN) is less than the actual cache line size. | |
171 | So set up the preconditions for L(nondcbz) and go there. */ | |
172 | .align 3 | |
173 | L(handletail32): | |
174 | clrrwi. rALIGN, rLEN, 5 | |
175 | b L(nondcbz) | |
176 | ||
177 | .align 5 | |
178 | L(small): | |
179 | /* Memset of 4 bytes or less. */ | |
180 | cmplwi cr5, rLEN, 1 | |
181 | cmplwi cr1, rLEN, 3 | |
182 | bltlr cr5 | |
183 | stb rCHR, 0(rMEMP) | |
184 | beqlr cr5 | |
185 | stb rCHR, 1(rMEMP) | |
186 | bltlr cr1 | |
187 | stb rCHR, 2(rMEMP) | |
188 | beqlr cr1 | |
189 | stb rCHR, 3(rMEMP) | |
190 | blr | |
191 | ||
192 | /* Memset of 0-31 bytes. */ | |
193 | .align 5 | |
194 | L(medium): | |
195 | cmplwi cr1, rLEN, 16 | |
196 | L(medium_tail2): | |
197 | add rMEMP, rMEMP, rLEN | |
198 | L(medium_tail): | |
199 | bt- 31, L(medium_31t) | |
200 | bt- 30, L(medium_30t) | |
201 | L(medium_30f): | |
202 | bt- 29, L(medium_29t) | |
203 | L(medium_29f): | |
204 | bge- cr1, L(medium_27t) | |
205 | bflr- 28 | |
206 | stw rCHR, -4(rMEMP) | |
207 | stw rCHR, -8(rMEMP) | |
208 | blr | |
209 | ||
210 | L(medium_31t): | |
211 | stbu rCHR, -1(rMEMP) | |
212 | bf- 30, L(medium_30f) | |
213 | L(medium_30t): | |
214 | sthu rCHR, -2(rMEMP) | |
215 | bf- 29, L(medium_29f) | |
216 | L(medium_29t): | |
217 | stwu rCHR, -4(rMEMP) | |
218 | blt- cr1, L(medium_27f) | |
219 | L(medium_27t): | |
220 | stw rCHR, -4(rMEMP) | |
221 | stw rCHR, -8(rMEMP) | |
222 | stw rCHR, -12(rMEMP) | |
223 | stwu rCHR, -16(rMEMP) | |
224 | L(medium_27f): | |
225 | bflr- 28 | |
226 | L(medium_28t): | |
227 | stw rCHR, -4(rMEMP) | |
228 | stw rCHR, -8(rMEMP) | |
229 | blr | |
230 | END (BP_SYM (memset)) | |
231 | libc_hidden_builtin_def (memset) |