1 /* SPDX-License-Identifier: GPL-2.0 */ <<
2 /* 1 /*
3 * arch/alpha/lib/ev6-stxncpy.S 2 * arch/alpha/lib/ev6-stxncpy.S
4 * 21264 version contributed by Rick Gorton <ri 3 * 21264 version contributed by Rick Gorton <rick.gorton@api-networks.com>
5 * 4 *
6 * Copy no more than COUNT bytes of the null-t 5 * Copy no more than COUNT bytes of the null-terminated string from
7 * SRC to DST. 6 * SRC to DST.
8 * 7 *
9 * This is an internal routine used by strncpy 8 * This is an internal routine used by strncpy, stpncpy, and strncat.
10 * As such, it uses special linkage convention 9 * As such, it uses special linkage conventions to make implementation
11 * of these public functions more efficient. 10 * of these public functions more efficient.
12 * 11 *
13 * On input: 12 * On input:
14 * t9 = return address 13 * t9 = return address
15 * a0 = DST 14 * a0 = DST
16 * a1 = SRC 15 * a1 = SRC
17 * a2 = COUNT 16 * a2 = COUNT
18 * 17 *
19 * Furthermore, COUNT may not be zero. 18 * Furthermore, COUNT may not be zero.
20 * 19 *
21 * On output: 20 * On output:
22 * t0 = last word written 21 * t0 = last word written
23 * t10 = bitmask (with one bit set) indic 22 * t10 = bitmask (with one bit set) indicating the byte position of
24 * the end of the range specified b 23 * the end of the range specified by COUNT
25 * t12 = bitmask (with one bit set) indic 24 * t12 = bitmask (with one bit set) indicating the last byte written
26 * a0 = unaligned address of the last *w 25 * a0 = unaligned address of the last *word* written
27 * a2 = the number of full words left in 26 * a2 = the number of full words left in COUNT
28 * 27 *
29 * Furthermore, v0, a3-a5, t11, and $at are un 28 * Furthermore, v0, a3-a5, t11, and $at are untouched.
30 * 29 *
31 * Much of the information about 21264 schedul 30 * Much of the information about 21264 scheduling/coding comes from:
32 * Compiler Writer's Guide for the Alpha 31 * Compiler Writer's Guide for the Alpha 21264
33 * abbreviated as 'CWG' in other comments 32 * abbreviated as 'CWG' in other comments here
34 * ftp.digital.com/pub/Digital/info/semic 33 * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
35 * Scheduling notation: 34 * Scheduling notation:
36 * E - either cluster 35 * E - either cluster
37 * U - upper subcluster; U0 - subcl 36 * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1
38 * L - lower subcluster; L0 - subcl 37 * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1
39 * Try not to change the actual algorithm if p 38 * Try not to change the actual algorithm if possible for consistency.
40 */ 39 */
41 40
42 #include <asm/regdef.h> 41 #include <asm/regdef.h>
43 42
44 .set noat 43 .set noat
45 .set noreorder 44 .set noreorder
46 45
47 .text 46 .text
48 47
49 /* There is a problem with either gdb (as of 4 48 /* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that
50 doesn't like putting the entry point for a 49 doesn't like putting the entry point for a procedure somewhere in the
51 middle of the procedure descriptor. Work a 50 middle of the procedure descriptor. Work around this by putting the
52 aligned copy in its own procedure descripto 51 aligned copy in its own procedure descriptor */
53 52
54 53
55 .ent stxncpy_aligned 54 .ent stxncpy_aligned
56 .align 4 55 .align 4
57 stxncpy_aligned: 56 stxncpy_aligned:
58 .frame sp, 0, t9, 0 57 .frame sp, 0, t9, 0
59 .prologue 0 58 .prologue 0
60 59
61 /* On entry to this basic block: 60 /* On entry to this basic block:
62 t0 == the first destination word fo 61 t0 == the first destination word for masking back in
63 t1 == the first source word. */ 62 t1 == the first source word. */
64 63
65 /* Create the 1st output word and dete 64 /* Create the 1st output word and detect 0's in the 1st input word. */
66 lda t2, -1 # E : build a 65 lda t2, -1 # E : build a mask against false zero
67 mskqh t2, a1, t2 # U : detect 66 mskqh t2, a1, t2 # U : detection in the src word (stall)
68 mskqh t1, a1, t3 # U : 67 mskqh t1, a1, t3 # U :
69 ornot t1, t2, t2 # E : (stall) 68 ornot t1, t2, t2 # E : (stall)
70 69
71 mskql t0, a1, t0 # U : assemble 70 mskql t0, a1, t0 # U : assemble the first output word
72 cmpbge zero, t2, t8 # E : bits set 71 cmpbge zero, t2, t8 # E : bits set iff null found
73 or t0, t3, t0 # E : (stall) 72 or t0, t3, t0 # E : (stall)
74 beq a2, $a_eoc # U : 73 beq a2, $a_eoc # U :
75 74
76 bne t8, $a_eos # U : 75 bne t8, $a_eos # U :
77 nop 76 nop
78 nop 77 nop
79 nop 78 nop
80 79
81 /* On entry to this basic block: 80 /* On entry to this basic block:
82 t0 == a source word not containing 81 t0 == a source word not containing a null. */
83 82
84 /* 83 /*
85 * nops here to: 84 * nops here to:
86 * separate store quads from load 85 * separate store quads from load quads
87 * limit of 1 bcond/quad to permi 86 * limit of 1 bcond/quad to permit training
88 */ 87 */
89 $a_loop: 88 $a_loop:
90 stq_u t0, 0(a0) # L : 89 stq_u t0, 0(a0) # L :
91 addq a0, 8, a0 # E : 90 addq a0, 8, a0 # E :
92 subq a2, 1, a2 # E : 91 subq a2, 1, a2 # E :
93 nop 92 nop
94 93
95 ldq_u t0, 0(a1) # L : 94 ldq_u t0, 0(a1) # L :
96 addq a1, 8, a1 # E : 95 addq a1, 8, a1 # E :
97 cmpbge zero, t0, t8 # E : 96 cmpbge zero, t0, t8 # E :
98 beq a2, $a_eoc # U : 97 beq a2, $a_eoc # U :
99 98
100 beq t8, $a_loop # U : 99 beq t8, $a_loop # U :
101 nop 100 nop
102 nop 101 nop
103 nop 102 nop
104 103
105 /* Take care of the final (partial) wo 104 /* Take care of the final (partial) word store. At this point
106 the end-of-count bit is set in t8 i 105 the end-of-count bit is set in t8 iff it applies.
107 106
108 On entry to this basic block we hav 107 On entry to this basic block we have:
109 t0 == the source word containing th 108 t0 == the source word containing the null
110 t8 == the cmpbge mask that found it 109 t8 == the cmpbge mask that found it. */
111 110
112 $a_eos: 111 $a_eos:
113 negq t8, t12 # E : find low 112 negq t8, t12 # E : find low bit set
114 and t8, t12, t12 # E : (stall) 113 and t8, t12, t12 # E : (stall)
115 /* For the sake of the cache, don't re 114 /* For the sake of the cache, don't read a destination word
116 if we're not going to need it. */ 115 if we're not going to need it. */
117 and t12, 0x80, t6 # E : (stall) 116 and t12, 0x80, t6 # E : (stall)
118 bne t6, 1f # U : (stall) 117 bne t6, 1f # U : (stall)
119 118
120 /* We're doing a partial word store an 119 /* We're doing a partial word store and so need to combine
121 our source and original destination 120 our source and original destination words. */
122 ldq_u t1, 0(a0) # L : 121 ldq_u t1, 0(a0) # L :
123 subq t12, 1, t6 # E : 122 subq t12, 1, t6 # E :
124 or t12, t6, t8 # E : (stall) 123 or t12, t6, t8 # E : (stall)
125 zapnot t0, t8, t0 # U : clear sr 124 zapnot t0, t8, t0 # U : clear src bytes > null (stall)
126 125
127 zap t1, t8, t1 # .. e1 : clea 126 zap t1, t8, t1 # .. e1 : clear dst bytes <= null
128 or t0, t1, t0 # e1 : (sta 127 or t0, t1, t0 # e1 : (stall)
129 nop 128 nop
130 nop 129 nop
131 130
132 1: stq_u t0, 0(a0) # L : 131 1: stq_u t0, 0(a0) # L :
133 ret (t9) # L0 : Latency 132 ret (t9) # L0 : Latency=3
134 nop 133 nop
135 nop 134 nop
136 135
137 /* Add the end-of-count bit to the eos 136 /* Add the end-of-count bit to the eos detection bitmask. */
138 $a_eoc: 137 $a_eoc:
139 or t10, t8, t8 # E : 138 or t10, t8, t8 # E :
140 br $a_eos # L0 : Latency 139 br $a_eos # L0 : Latency=3
141 nop 140 nop
142 nop 141 nop
143 142
144 .end stxncpy_aligned 143 .end stxncpy_aligned
145 144
146 .align 4 145 .align 4
147 .ent __stxncpy 146 .ent __stxncpy
148 .globl __stxncpy 147 .globl __stxncpy
149 __stxncpy: 148 __stxncpy:
150 .frame sp, 0, t9, 0 149 .frame sp, 0, t9, 0
151 .prologue 0 150 .prologue 0
152 151
153 /* Are source and destination co-align 152 /* Are source and destination co-aligned? */
154 xor a0, a1, t1 # E : 153 xor a0, a1, t1 # E :
155 and a0, 7, t0 # E : find des 154 and a0, 7, t0 # E : find dest misalignment
156 and t1, 7, t1 # E : (stall) 155 and t1, 7, t1 # E : (stall)
157 addq a2, t0, a2 # E : bias cou 156 addq a2, t0, a2 # E : bias count by dest misalignment (stall)
158 157
159 subq a2, 1, a2 # E : 158 subq a2, 1, a2 # E :
160 and a2, 7, t2 # E : (stall) 159 and a2, 7, t2 # E : (stall)
161 srl a2, 3, a2 # U : a2 = loo 160 srl a2, 3, a2 # U : a2 = loop counter = (count - 1)/8 (stall)
162 addq zero, 1, t10 # E : 161 addq zero, 1, t10 # E :
163 162
164 sll t10, t2, t10 # U : t10 = bi 163 sll t10, t2, t10 # U : t10 = bitmask of last count byte
165 bne t1, $unaligned # U : 164 bne t1, $unaligned # U :
166 /* We are co-aligned; take care of a p 165 /* We are co-aligned; take care of a partial first word. */
167 ldq_u t1, 0(a1) # L : load fir 166 ldq_u t1, 0(a1) # L : load first src word
168 addq a1, 8, a1 # E : 167 addq a1, 8, a1 # E :
169 168
170 beq t0, stxncpy_aligned # U : 169 beq t0, stxncpy_aligned # U : avoid loading dest word if not needed
171 ldq_u t0, 0(a0) # L : 170 ldq_u t0, 0(a0) # L :
172 nop 171 nop
173 nop 172 nop
174 173
175 br stxncpy_aligned # .. e1 : 174 br stxncpy_aligned # .. e1 :
176 nop 175 nop
177 nop 176 nop
178 nop 177 nop
179 178
180 179
181 180
182 /* The source and destination are not co-align 181 /* The source and destination are not co-aligned. Align the destination
183 and cope. We have to be very careful about 182 and cope. We have to be very careful about not reading too much and
184 causing a SEGV. */ 183 causing a SEGV. */
185 184
186 .align 4 185 .align 4
187 $u_head: 186 $u_head:
188 /* We know just enough now to be able 187 /* We know just enough now to be able to assemble the first
189 full source word. We can still fin 188 full source word. We can still find a zero at the end of it
190 that prevents us from outputting th 189 that prevents us from outputting the whole thing.
191 190
192 On entry to this basic block: 191 On entry to this basic block:
193 t0 == the first dest word, unmasked 192 t0 == the first dest word, unmasked
194 t1 == the shifted low bits of the f 193 t1 == the shifted low bits of the first source word
195 t6 == bytemask that is -1 in dest w 194 t6 == bytemask that is -1 in dest word bytes */
196 195
197 ldq_u t2, 8(a1) # L : Latency= 196 ldq_u t2, 8(a1) # L : Latency=3 load second src word
198 addq a1, 8, a1 # E : 197 addq a1, 8, a1 # E :
199 mskql t0, a0, t0 # U : mask tra 198 mskql t0, a0, t0 # U : mask trailing garbage in dst
200 extqh t2, a1, t4 # U : (3 cycle 199 extqh t2, a1, t4 # U : (3 cycle stall on t2)
201 200
202 or t1, t4, t1 # E : first al 201 or t1, t4, t1 # E : first aligned src word complete (stall)
203 mskqh t1, a0, t1 # U : mask lea 202 mskqh t1, a0, t1 # U : mask leading garbage in src (stall)
204 or t0, t1, t0 # E : first ou 203 or t0, t1, t0 # E : first output word complete (stall)
205 or t0, t6, t6 # E : mask ori 204 or t0, t6, t6 # E : mask original data for zero test (stall)
206 205
207 cmpbge zero, t6, t8 # E : 206 cmpbge zero, t6, t8 # E :
208 beq a2, $u_eocfin # U : 207 beq a2, $u_eocfin # U :
209 lda t6, -1 # E : 208 lda t6, -1 # E :
210 nop 209 nop
211 210
212 bne t8, $u_final # U : 211 bne t8, $u_final # U :
213 mskql t6, a1, t6 # U : mask out 212 mskql t6, a1, t6 # U : mask out bits already seen
214 stq_u t0, 0(a0) # L : store fi 213 stq_u t0, 0(a0) # L : store first output word
215 or t6, t2, t2 # E : (stall) 214 or t6, t2, t2 # E : (stall)
216 215
217 cmpbge zero, t2, t8 # E : find nul 216 cmpbge zero, t2, t8 # E : find nulls in second partial
218 addq a0, 8, a0 # E : 217 addq a0, 8, a0 # E :
219 subq a2, 1, a2 # E : 218 subq a2, 1, a2 # E :
220 bne t8, $u_late_head_exit # U : 219 bne t8, $u_late_head_exit # U :
221 220
222 /* Finally, we've got all the stupid l 221 /* Finally, we've got all the stupid leading edge cases taken care
223 of and we can set up to enter the m 222 of and we can set up to enter the main loop. */
224 extql t2, a1, t1 # U : position 223 extql t2, a1, t1 # U : position hi-bits of lo word
225 beq a2, $u_eoc # U : 224 beq a2, $u_eoc # U :
226 ldq_u t2, 8(a1) # L : read nex 225 ldq_u t2, 8(a1) # L : read next high-order source word
227 addq a1, 8, a1 # E : 226 addq a1, 8, a1 # E :
228 227
229 extqh t2, a1, t0 # U : position 228 extqh t2, a1, t0 # U : position lo-bits of hi word (stall)
230 cmpbge zero, t2, t8 # E : 229 cmpbge zero, t2, t8 # E :
231 nop 230 nop
232 bne t8, $u_eos # U : 231 bne t8, $u_eos # U :
233 232
234 /* Unaligned copy main loop. In order 233 /* Unaligned copy main loop. In order to avoid reading too much,
235 the loop is structured to detect ze 234 the loop is structured to detect zeros in aligned source words.
236 This has, unfortunately, effectivel 235 This has, unfortunately, effectively pulled half of a loop
237 iteration out into the head and hal 236 iteration out into the head and half into the tail, but it does
238 prevent nastiness from accumulating 237 prevent nastiness from accumulating in the very thing we want
239 to run as fast as possible. 238 to run as fast as possible.
240 239
241 On entry to this basic block: 240 On entry to this basic block:
242 t0 == the shifted low-order bits fr 241 t0 == the shifted low-order bits from the current source word
243 t1 == the shifted high-order bits f 242 t1 == the shifted high-order bits from the previous source word
244 t2 == the unshifted current source 243 t2 == the unshifted current source word
245 244
246 We further know that t2 does not co 245 We further know that t2 does not contain a null terminator. */
247 246
248 .align 4 247 .align 4
249 $u_loop: 248 $u_loop:
250 or t0, t1, t0 # E : current 249 or t0, t1, t0 # E : current dst word now complete
251 subq a2, 1, a2 # E : decremen 250 subq a2, 1, a2 # E : decrement word count
252 extql t2, a1, t1 # U : extract 251 extql t2, a1, t1 # U : extract low bits for next time
253 addq a0, 8, a0 # E : 252 addq a0, 8, a0 # E :
254 253
255 stq_u t0, -8(a0) # U : save the 254 stq_u t0, -8(a0) # U : save the current word
256 beq a2, $u_eoc # U : 255 beq a2, $u_eoc # U :
257 ldq_u t2, 8(a1) # U : Latency= 256 ldq_u t2, 8(a1) # U : Latency=3 load high word for next time
258 addq a1, 8, a1 # E : 257 addq a1, 8, a1 # E :
259 258
260 extqh t2, a1, t0 # U : extract 259 extqh t2, a1, t0 # U : extract low bits (2 cycle stall)
261 cmpbge zero, t2, t8 # E : test new 260 cmpbge zero, t2, t8 # E : test new word for eos
262 nop 261 nop
263 beq t8, $u_loop # U : 262 beq t8, $u_loop # U :
264 263
265 /* We've found a zero somewhere in the 264 /* We've found a zero somewhere in the source word we just read.
266 If it resides in the lower half, we 265 If it resides in the lower half, we have one (probably partial)
267 word to write out, and if it reside 266 word to write out, and if it resides in the upper half, we
268 have one full and one partial word 267 have one full and one partial word left to write out.
269 268
270 On entry to this basic block: 269 On entry to this basic block:
271 t0 == the shifted low-order bits fr 270 t0 == the shifted low-order bits from the current source word
272 t1 == the shifted high-order bits f 271 t1 == the shifted high-order bits from the previous source word
273 t2 == the unshifted current source 272 t2 == the unshifted current source word. */
274 $u_eos: 273 $u_eos:
275 or t0, t1, t0 # E : first (p 274 or t0, t1, t0 # E : first (partial) source word complete
276 nop 275 nop
277 cmpbge zero, t0, t8 # E : is the n 276 cmpbge zero, t0, t8 # E : is the null in this first bit? (stall)
278 bne t8, $u_final # U : (stall) 277 bne t8, $u_final # U : (stall)
279 278
280 stq_u t0, 0(a0) # L : the null 279 stq_u t0, 0(a0) # L : the null was in the high-order bits
281 addq a0, 8, a0 # E : 280 addq a0, 8, a0 # E :
282 subq a2, 1, a2 # E : 281 subq a2, 1, a2 # E :
283 nop 282 nop
284 283
285 $u_late_head_exit: 284 $u_late_head_exit:
286 extql t2, a1, t0 # U : 285 extql t2, a1, t0 # U :
287 cmpbge zero, t0, t8 # E : 286 cmpbge zero, t0, t8 # E :
288 or t8, t10, t6 # E : (stall) 287 or t8, t10, t6 # E : (stall)
289 cmoveq a2, t6, t8 # E : Latency= 288 cmoveq a2, t6, t8 # E : Latency=2, extra map slot (stall)
290 289
291 /* Take care of a final (probably part 290 /* Take care of a final (probably partial) result word.
292 On entry to this basic block: 291 On entry to this basic block:
293 t0 == assembled source word 292 t0 == assembled source word
294 t8 == cmpbge mask that found the nu 293 t8 == cmpbge mask that found the null. */
295 $u_final: 294 $u_final:
296 negq t8, t6 # E : isolate 295 negq t8, t6 # E : isolate low bit set
297 and t6, t8, t12 # E : (stall) 296 and t6, t8, t12 # E : (stall)
298 and t12, 0x80, t6 # E : avoid de 297 and t12, 0x80, t6 # E : avoid dest word load if we can (stall)
299 bne t6, 1f # U : (stall) 298 bne t6, 1f # U : (stall)
300 299
301 ldq_u t1, 0(a0) # L : 300 ldq_u t1, 0(a0) # L :
302 subq t12, 1, t6 # E : 301 subq t12, 1, t6 # E :
303 or t6, t12, t8 # E : (stall) 302 or t6, t12, t8 # E : (stall)
304 zapnot t0, t8, t0 # U : kill sou 303 zapnot t0, t8, t0 # U : kill source bytes > null
305 304
306 zap t1, t8, t1 # U : kill des 305 zap t1, t8, t1 # U : kill dest bytes <= null
307 or t0, t1, t0 # E : (stall) 306 or t0, t1, t0 # E : (stall)
308 nop 307 nop
309 nop 308 nop
310 309
311 1: stq_u t0, 0(a0) # L : 310 1: stq_u t0, 0(a0) # L :
312 ret (t9) # L0 : Latency 311 ret (t9) # L0 : Latency=3
313 312
314 /* Got to end-of-count before end of 313 /* Got to end-of-count before end of string.
315 On entry to this basic block: 314 On entry to this basic block:
316 t1 == the shifted high-order bits 315 t1 == the shifted high-order bits from the previous source word */
317 $u_eoc: 316 $u_eoc:
318 and a1, 7, t6 # E : avoid fi 317 and a1, 7, t6 # E : avoid final load if possible
319 sll t10, t6, t6 # U : (stall) 318 sll t10, t6, t6 # U : (stall)
320 and t6, 0xff, t6 # E : (stall) 319 and t6, 0xff, t6 # E : (stall)
321 bne t6, 1f # U : (stall) 320 bne t6, 1f # U : (stall)
322 321
323 ldq_u t2, 8(a1) # L : load fin 322 ldq_u t2, 8(a1) # L : load final src word
324 nop 323 nop
325 extqh t2, a1, t0 # U : extract 324 extqh t2, a1, t0 # U : extract low bits for last word (stall)
326 or t1, t0, t1 # E : (stall) 325 or t1, t0, t1 # E : (stall)
327 326
328 1: cmpbge zero, t1, t8 # E : 327 1: cmpbge zero, t1, t8 # E :
329 mov t1, t0 # E : 328 mov t1, t0 # E :
330 329
331 $u_eocfin: # end-of-count 330 $u_eocfin: # end-of-count, final word
332 or t10, t8, t8 # E : 331 or t10, t8, t8 # E :
333 br $u_final # L0 : Latency 332 br $u_final # L0 : Latency=3
334 333
335 /* Unaligned copy entry point. */ 334 /* Unaligned copy entry point. */
336 .align 4 335 .align 4
337 $unaligned: 336 $unaligned:
338 337
339 ldq_u t1, 0(a1) # L : load fir 338 ldq_u t1, 0(a1) # L : load first source word
340 and a0, 7, t4 # E : find des 339 and a0, 7, t4 # E : find dest misalignment
341 and a1, 7, t5 # E : find src 340 and a1, 7, t5 # E : find src misalignment
342 /* Conditionally load the first destin 341 /* Conditionally load the first destination word and a bytemask
343 with 0xff indicating that the desti 342 with 0xff indicating that the destination byte is sacrosanct. */
344 mov zero, t0 # E : 343 mov zero, t0 # E :
345 344
346 mov zero, t6 # E : 345 mov zero, t6 # E :
347 beq t4, 1f # U : 346 beq t4, 1f # U :
348 ldq_u t0, 0(a0) # L : 347 ldq_u t0, 0(a0) # L :
349 lda t6, -1 # E : 348 lda t6, -1 # E :
350 349
351 mskql t6, a0, t6 # U : 350 mskql t6, a0, t6 # U :
352 nop 351 nop
353 nop 352 nop
354 subq a1, t4, a1 # E : sub dest 353 subq a1, t4, a1 # E : sub dest misalignment from src addr
355 354
356 /* If source misalignment is larger th 355 /* If source misalignment is larger than dest misalignment, we need
357 extra startup checks to avoid SEGV. 356 extra startup checks to avoid SEGV. */
358 357
359 1: cmplt t4, t5, t12 # E : 358 1: cmplt t4, t5, t12 # E :
360 extql t1, a1, t1 # U : shift sr 359 extql t1, a1, t1 # U : shift src into place
361 lda t2, -1 # E : for crea 360 lda t2, -1 # E : for creating masks later
362 beq t12, $u_head # U : (stall) 361 beq t12, $u_head # U : (stall)
363 362
364 extql t2, a1, t2 # U : 363 extql t2, a1, t2 # U :
365 cmpbge zero, t1, t8 # E : is there 364 cmpbge zero, t1, t8 # E : is there a zero?
366 andnot t2, t6, t2 # E : dest mas !! 365 andnot t2, t6, t12 # E : dest mask for a single word copy
367 or t8, t10, t5 # E : test for 366 or t8, t10, t5 # E : test for end-of-count too
368 367
369 cmpbge zero, t2, t3 # E : 368 cmpbge zero, t2, t3 # E :
370 cmoveq a2, t5, t8 # E : Latency= 369 cmoveq a2, t5, t8 # E : Latency=2, extra map slot
371 nop # E : keep wit 370 nop # E : keep with cmoveq
372 andnot t8, t3, t8 # E : (stall) 371 andnot t8, t3, t8 # E : (stall)
373 372
374 beq t8, $u_head # U : 373 beq t8, $u_head # U :
375 /* At this point we've found a zero in 374 /* At this point we've found a zero in the first partial word of
376 the source. We need to isolate the 375 the source. We need to isolate the valid source data and mask
377 it into the original destination da 376 it into the original destination data. (Incidentally, we know
378 that we'll need at least one byte o 377 that we'll need at least one byte of that original dest word.) */
379 ldq_u t0, 0(a0) # L : 378 ldq_u t0, 0(a0) # L :
380 negq t8, t6 # E : build bi 379 negq t8, t6 # E : build bitmask of bytes <= zero
381 mskqh t1, t4, t1 # U : 380 mskqh t1, t4, t1 # U :
382 381
383 and t6, t8, t12 # E : !! 382 and t6, t8, t2 # E :
384 subq t12, 1, t6 # E : (stall) !! 383 subq t2, 1, t6 # E : (stall)
385 or t6, t12, t8 # E : (stall) !! 384 or t6, t2, t8 # E : (stall)
386 zapnot t2, t8, t2 # U : prepare !! 385 zapnot t12, t8, t12 # U : prepare source word; mirror changes (stall)
387 386
388 zapnot t1, t8, t1 # U : to sourc 387 zapnot t1, t8, t1 # U : to source validity mask
389 andnot t0, t2, t0 # E : zero pla !! 388 andnot t0, t12, t0 # E : zero place for source to reside
390 or t0, t1, t0 # E : and put 389 or t0, t1, t0 # E : and put it there (stall both t0, t1)
391 stq_u t0, 0(a0) # L : (stall) 390 stq_u t0, 0(a0) # L : (stall)
392 391
393 ret (t9) # L0 : Latency 392 ret (t9) # L0 : Latency=3
394 nop 393 nop
395 nop 394 nop
396 nop 395 nop
397 396
398 .end __stxncpy 397 .end __stxncpy
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.