diff options
author | David S. Miller <davem@sunset.davemloft.net> | 2007-10-02 04:03:09 -0400 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2007-10-02 04:03:09 -0400 |
commit | 25e5566ed38650f7990041fcd20571d6ddd2a040 (patch) | |
tree | 26d5bf006a6945930102fa01ad0edcd479b39bbf /arch/sparc64/lib/NGmemcpy.S | |
parent | 8cc8c28a9acdceda0e60519167a052cc3408c5c3 (diff) |
[SPARC64]: Fix missing load-twin usage in Niagara-1 memcpy.
For the case where the source is not aligned modulo 8
we don't use load-twins to suck the data in and this
kills performance since normal loads allocate in the
L1 cache (unlike load-twin) and thus big memcpys swipe
the entire L1 D-cache.
We need to allocate a register window to implement this
properly, but that actually simplifies a lot of things
as a nice side-effect.
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc64/lib/NGmemcpy.S')
-rw-r--r-- | arch/sparc64/lib/NGmemcpy.S | 371 |
1 files changed, 213 insertions, 158 deletions
diff --git a/arch/sparc64/lib/NGmemcpy.S b/arch/sparc64/lib/NGmemcpy.S index 66063a9a66b8..605cb3f09900 100644 --- a/arch/sparc64/lib/NGmemcpy.S +++ b/arch/sparc64/lib/NGmemcpy.S | |||
@@ -1,6 +1,6 @@ | |||
1 | /* NGmemcpy.S: Niagara optimized memcpy. | 1 | /* NGmemcpy.S: Niagara optimized memcpy. |
2 | * | 2 | * |
3 | * Copyright (C) 2006 David S. Miller (davem@davemloft.net) | 3 | * Copyright (C) 2006, 2007 David S. Miller (davem@davemloft.net) |
4 | */ | 4 | */ |
5 | 5 | ||
6 | #ifdef __KERNEL__ | 6 | #ifdef __KERNEL__ |
@@ -16,6 +16,12 @@ | |||
16 | wr %g0, ASI_PNF, %asi | 16 | wr %g0, ASI_PNF, %asi |
17 | #endif | 17 | #endif |
18 | 18 | ||
19 | #ifdef __sparc_v9__ | ||
20 | #define SAVE_AMOUNT 128 | ||
21 | #else | ||
22 | #define SAVE_AMOUNT 64 | ||
23 | #endif | ||
24 | |||
19 | #ifndef STORE_ASI | 25 | #ifndef STORE_ASI |
20 | #define STORE_ASI ASI_BLK_INIT_QUAD_LDD_P | 26 | #define STORE_ASI ASI_BLK_INIT_QUAD_LDD_P |
21 | #endif | 27 | #endif |
@@ -50,7 +56,11 @@ | |||
50 | #endif | 56 | #endif |
51 | 57 | ||
52 | #ifndef STORE_INIT | 58 | #ifndef STORE_INIT |
59 | #ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA | ||
53 | #define STORE_INIT(src,addr) stxa src, [addr] %asi | 60 | #define STORE_INIT(src,addr) stxa src, [addr] %asi |
61 | #else | ||
62 | #define STORE_INIT(src,addr) stx src, [addr + 0x00] | ||
63 | #endif | ||
54 | #endif | 64 | #endif |
55 | 65 | ||
56 | #ifndef FUNC_NAME | 66 | #ifndef FUNC_NAME |
@@ -73,18 +83,19 @@ | |||
73 | 83 | ||
74 | .globl FUNC_NAME | 84 | .globl FUNC_NAME |
75 | .type FUNC_NAME,#function | 85 | .type FUNC_NAME,#function |
76 | FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ | 86 | FUNC_NAME: /* %i0=dst, %i1=src, %i2=len */ |
77 | srlx %o2, 31, %g2 | 87 | PREAMBLE |
88 | save %sp, -SAVE_AMOUNT, %sp | ||
89 | srlx %i2, 31, %g2 | ||
78 | cmp %g2, 0 | 90 | cmp %g2, 0 |
79 | tne %xcc, 5 | 91 | tne %xcc, 5 |
80 | PREAMBLE | 92 | mov %i0, %o0 |
81 | mov %o0, GLOBAL_SPARE | 93 | cmp %i2, 0 |
82 | cmp %o2, 0 | ||
83 | be,pn %XCC, 85f | 94 | be,pn %XCC, 85f |
84 | or %o0, %o1, %o3 | 95 | or %o0, %i1, %i3 |
85 | cmp %o2, 16 | 96 | cmp %i2, 16 |
86 | blu,a,pn %XCC, 80f | 97 | blu,a,pn %XCC, 80f |
87 | or %o3, %o2, %o3 | 98 | or %i3, %i2, %i3 |
88 | 99 | ||
89 | /* 2 blocks (128 bytes) is the minimum we can do the block | 100 | /* 2 blocks (128 bytes) is the minimum we can do the block |
90 | * copy with. We need to ensure that we'll iterate at least | 101 | * copy with. We need to ensure that we'll iterate at least |
@@ -93,31 +104,31 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ | |||
93 | * to (64 - 1) bytes from the length before we perform the | 104 | * to (64 - 1) bytes from the length before we perform the |
94 | * block copy loop. | 105 | * block copy loop. |
95 | */ | 106 | */ |
96 | cmp %o2, (2 * 64) | 107 | cmp %i2, (2 * 64) |
97 | blu,pt %XCC, 70f | 108 | blu,pt %XCC, 70f |
98 | andcc %o3, 0x7, %g0 | 109 | andcc %i3, 0x7, %g0 |
99 | 110 | ||
100 | /* %o0: dst | 111 | /* %o0: dst |
101 | * %o1: src | 112 | * %i1: src |
102 | * %o2: len (known to be >= 128) | 113 | * %i2: len (known to be >= 128) |
103 | * | 114 | * |
104 | * The block copy loops will use %o4/%o5,%g2/%g3 as | 115 | * The block copy loops will use %i4/%i5,%g2/%g3 as |
105 | * temporaries while copying the data. | 116 | * temporaries while copying the data. |
106 | */ | 117 | */ |
107 | 118 | ||
108 | LOAD(prefetch, %o1, #one_read) | 119 | LOAD(prefetch, %i1, #one_read) |
109 | wr %g0, STORE_ASI, %asi | 120 | wr %g0, STORE_ASI, %asi |
110 | 121 | ||
111 | /* Align destination on 64-byte boundary. */ | 122 | /* Align destination on 64-byte boundary. */ |
112 | andcc %o0, (64 - 1), %o4 | 123 | andcc %o0, (64 - 1), %i4 |
113 | be,pt %XCC, 2f | 124 | be,pt %XCC, 2f |
114 | sub %o4, 64, %o4 | 125 | sub %i4, 64, %i4 |
115 | sub %g0, %o4, %o4 ! bytes to align dst | 126 | sub %g0, %i4, %i4 ! bytes to align dst |
116 | sub %o2, %o4, %o2 | 127 | sub %i2, %i4, %i2 |
117 | 1: subcc %o4, 1, %o4 | 128 | 1: subcc %i4, 1, %i4 |
118 | EX_LD(LOAD(ldub, %o1, %g1)) | 129 | EX_LD(LOAD(ldub, %i1, %g1)) |
119 | EX_ST(STORE(stb, %g1, %o0)) | 130 | EX_ST(STORE(stb, %g1, %o0)) |
120 | add %o1, 1, %o1 | 131 | add %i1, 1, %i1 |
121 | bne,pt %XCC, 1b | 132 | bne,pt %XCC, 1b |
122 | add %o0, 1, %o0 | 133 | add %o0, 1, %o0 |
123 | 134 | ||
@@ -136,111 +147,155 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ | |||
136 | * aligned store data at a time, this is easy to ensure. | 147 | * aligned store data at a time, this is easy to ensure. |
137 | */ | 148 | */ |
138 | 2: | 149 | 2: |
139 | andcc %o1, (16 - 1), %o4 | 150 | andcc %i1, (16 - 1), %i4 |
140 | andn %o2, (64 - 1), %g1 ! block copy loop iterator | 151 | andn %i2, (64 - 1), %g1 ! block copy loop iterator |
141 | sub %o2, %g1, %o2 ! final sub-block copy bytes | ||
142 | be,pt %XCC, 50f | 152 | be,pt %XCC, 50f |
143 | cmp %o4, 8 | 153 | sub %i2, %g1, %i2 ! final sub-block copy bytes |
144 | be,a,pt %XCC, 10f | 154 | |
145 | sub %o1, 0x8, %o1 | 155 | cmp %i4, 8 |
156 | be,pt %XCC, 10f | ||
157 | sub %i1, %i4, %i1 | ||
146 | 158 | ||
147 | /* Neither 8-byte nor 16-byte aligned, shift and mask. */ | 159 | /* Neither 8-byte nor 16-byte aligned, shift and mask. */ |
148 | mov %g1, %o4 | 160 | and %i4, 0x7, GLOBAL_SPARE |
149 | and %o1, 0x7, %g1 | 161 | sll GLOBAL_SPARE, 3, GLOBAL_SPARE |
150 | sll %g1, 3, %g1 | 162 | mov 64, %i5 |
151 | mov 64, %o3 | 163 | EX_LD(LOAD_TWIN(%i1, %g2, %g3)) |
152 | andn %o1, 0x7, %o1 | 164 | sub %i5, GLOBAL_SPARE, %i5 |
153 | EX_LD(LOAD(ldx, %o1, %g2)) | 165 | mov 16, %o4 |
154 | sub %o3, %g1, %o3 | 166 | mov 32, %o5 |
155 | sllx %g2, %g1, %g2 | 167 | mov 48, %o7 |
168 | mov 64, %i3 | ||
169 | |||
170 | bg,pn %XCC, 9f | ||
171 | nop | ||
156 | 172 | ||
157 | #define SWIVEL_ONE_DWORD(SRC, TMP1, TMP2, PRE_VAL, PRE_SHIFT, POST_SHIFT, DST)\ | 173 | #define MIX_THREE_WORDS(WORD1, WORD2, WORD3, PRE_SHIFT, POST_SHIFT, TMP) \ |
158 | EX_LD(LOAD(ldx, SRC, TMP1)); \ | 174 | sllx WORD1, POST_SHIFT, WORD1; \ |
159 | srlx TMP1, PRE_SHIFT, TMP2; \ | 175 | srlx WORD2, PRE_SHIFT, TMP; \ |
160 | or TMP2, PRE_VAL, TMP2; \ | 176 | sllx WORD2, POST_SHIFT, WORD2; \ |
161 | EX_ST(STORE_INIT(TMP2, DST)); \ | 177 | or WORD1, TMP, WORD1; \ |
162 | sllx TMP1, POST_SHIFT, PRE_VAL; | 178 | srlx WORD3, PRE_SHIFT, TMP; \ |
163 | 179 | or WORD2, TMP, WORD2; | |
164 | 1: add %o1, 0x8, %o1 | 180 | |
165 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x00) | 181 | 8: EX_LD(LOAD_TWIN(%i1 + %o4, %o2, %o3)) |
166 | add %o1, 0x8, %o1 | 182 | MIX_THREE_WORDS(%g2, %g3, %o2, %i5, GLOBAL_SPARE, %o1) |
167 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x08) | 183 | LOAD(prefetch, %i1 + %i3, #one_read) |
168 | add %o1, 0x8, %o1 | 184 | |
169 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x10) | 185 | EX_ST(STORE_INIT(%g2, %o0 + 0x00)) |
170 | add %o1, 0x8, %o1 | 186 | EX_ST(STORE_INIT(%g3, %o0 + 0x08)) |
171 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x18) | 187 | |
172 | add %o1, 32, %o1 | 188 | EX_LD(LOAD_TWIN(%i1 + %o5, %g2, %g3)) |
173 | LOAD(prefetch, %o1, #one_read) | 189 | MIX_THREE_WORDS(%o2, %o3, %g2, %i5, GLOBAL_SPARE, %o1) |
174 | sub %o1, 32 - 8, %o1 | 190 | |
175 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x20) | 191 | EX_ST(STORE_INIT(%o2, %o0 + 0x10)) |
176 | add %o1, 8, %o1 | 192 | EX_ST(STORE_INIT(%o3, %o0 + 0x18)) |
177 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x28) | 193 | |
178 | add %o1, 8, %o1 | 194 | EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) |
179 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x30) | 195 | MIX_THREE_WORDS(%g2, %g3, %o2, %i5, GLOBAL_SPARE, %o1) |
180 | add %o1, 8, %o1 | 196 | |
181 | SWIVEL_ONE_DWORD(%o1, %g3, %o5, %g2, %o3, %g1, %o0 + 0x38) | 197 | EX_ST(STORE_INIT(%g2, %o0 + 0x20)) |
182 | subcc %o4, 64, %o4 | 198 | EX_ST(STORE_INIT(%g3, %o0 + 0x28)) |
183 | bne,pt %XCC, 1b | 199 | |
200 | EX_LD(LOAD_TWIN(%i1 + %i3, %g2, %g3)) | ||
201 | add %i1, 64, %i1 | ||
202 | MIX_THREE_WORDS(%o2, %o3, %g2, %i5, GLOBAL_SPARE, %o1) | ||
203 | |||
204 | EX_ST(STORE_INIT(%o2, %o0 + 0x30)) | ||
205 | EX_ST(STORE_INIT(%o3, %o0 + 0x38)) | ||
206 | |||
207 | subcc %g1, 64, %g1 | ||
208 | bne,pt %XCC, 8b | ||
184 | add %o0, 64, %o0 | 209 | add %o0, 64, %o0 |
185 | 210 | ||
186 | #undef SWIVEL_ONE_DWORD | 211 | ba,pt %XCC, 60f |
212 | add %i1, %i4, %i1 | ||
213 | |||
214 | 9: EX_LD(LOAD_TWIN(%i1 + %o4, %o2, %o3)) | ||
215 | MIX_THREE_WORDS(%g3, %o2, %o3, %i5, GLOBAL_SPARE, %o1) | ||
216 | LOAD(prefetch, %i1 + %i3, #one_read) | ||
217 | |||
218 | EX_ST(STORE_INIT(%g3, %o0 + 0x00)) | ||
219 | EX_ST(STORE_INIT(%o2, %o0 + 0x08)) | ||
220 | |||
221 | EX_LD(LOAD_TWIN(%i1 + %o5, %g2, %g3)) | ||
222 | MIX_THREE_WORDS(%o3, %g2, %g3, %i5, GLOBAL_SPARE, %o1) | ||
223 | |||
224 | EX_ST(STORE_INIT(%o3, %o0 + 0x10)) | ||
225 | EX_ST(STORE_INIT(%g2, %o0 + 0x18)) | ||
226 | |||
227 | EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) | ||
228 | MIX_THREE_WORDS(%g3, %o2, %o3, %i5, GLOBAL_SPARE, %o1) | ||
229 | |||
230 | EX_ST(STORE_INIT(%g3, %o0 + 0x20)) | ||
231 | EX_ST(STORE_INIT(%o2, %o0 + 0x28)) | ||
232 | |||
233 | EX_LD(LOAD_TWIN(%i1 + %i3, %g2, %g3)) | ||
234 | add %i1, 64, %i1 | ||
235 | MIX_THREE_WORDS(%o3, %g2, %g3, %i5, GLOBAL_SPARE, %o1) | ||
236 | |||
237 | EX_ST(STORE_INIT(%o3, %o0 + 0x30)) | ||
238 | EX_ST(STORE_INIT(%g2, %o0 + 0x38)) | ||
239 | |||
240 | subcc %g1, 64, %g1 | ||
241 | bne,pt %XCC, 9b | ||
242 | add %o0, 64, %o0 | ||
187 | 243 | ||
188 | srl %g1, 3, %g1 | ||
189 | ba,pt %XCC, 60f | 244 | ba,pt %XCC, 60f |
190 | add %o1, %g1, %o1 | 245 | add %i1, %i4, %i1 |
191 | 246 | ||
192 | 10: /* Destination is 64-byte aligned, source was only 8-byte | 247 | 10: /* Destination is 64-byte aligned, source was only 8-byte |
193 | * aligned but it has been subtracted by 8 and we perform | 248 | * aligned but it has been subtracted by 8 and we perform |
194 | * one twin load ahead, then add 8 back into source when | 249 | * one twin load ahead, then add 8 back into source when |
195 | * we finish the loop. | 250 | * we finish the loop. |
196 | */ | 251 | */ |
197 | EX_LD(LOAD_TWIN(%o1, %o4, %o5)) | 252 | EX_LD(LOAD_TWIN(%i1, %o4, %o5)) |
198 | 1: add %o1, 16, %o1 | 253 | mov 16, %o7 |
199 | EX_LD(LOAD_TWIN(%o1, %g2, %g3)) | 254 | mov 32, %g2 |
200 | add %o1, 16 + 32, %o1 | 255 | mov 48, %g3 |
201 | LOAD(prefetch, %o1, #one_read) | 256 | mov 64, %o1 |
202 | sub %o1, 32, %o1 | 257 | 1: EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) |
258 | LOAD(prefetch, %i1 + %o1, #one_read) | ||
203 | EX_ST(STORE_INIT(%o5, %o0 + 0x00)) ! initializes cache line | 259 | EX_ST(STORE_INIT(%o5, %o0 + 0x00)) ! initializes cache line |
204 | EX_ST(STORE_INIT(%g2, %o0 + 0x08)) | 260 | EX_ST(STORE_INIT(%o2, %o0 + 0x08)) |
205 | EX_LD(LOAD_TWIN(%o1, %o4, %o5)) | 261 | EX_LD(LOAD_TWIN(%i1 + %g2, %o4, %o5)) |
206 | add %o1, 16, %o1 | 262 | EX_ST(STORE_INIT(%o3, %o0 + 0x10)) |
207 | EX_ST(STORE_INIT(%g3, %o0 + 0x10)) | ||
208 | EX_ST(STORE_INIT(%o4, %o0 + 0x18)) | 263 | EX_ST(STORE_INIT(%o4, %o0 + 0x18)) |
209 | EX_LD(LOAD_TWIN(%o1, %g2, %g3)) | 264 | EX_LD(LOAD_TWIN(%i1 + %g3, %o2, %o3)) |
210 | add %o1, 16, %o1 | ||
211 | EX_ST(STORE_INIT(%o5, %o0 + 0x20)) | 265 | EX_ST(STORE_INIT(%o5, %o0 + 0x20)) |
212 | EX_ST(STORE_INIT(%g2, %o0 + 0x28)) | 266 | EX_ST(STORE_INIT(%o2, %o0 + 0x28)) |
213 | EX_LD(LOAD_TWIN(%o1, %o4, %o5)) | 267 | EX_LD(LOAD_TWIN(%i1 + %o1, %o4, %o5)) |
214 | EX_ST(STORE_INIT(%g3, %o0 + 0x30)) | 268 | add %i1, 64, %i1 |
269 | EX_ST(STORE_INIT(%o3, %o0 + 0x30)) | ||
215 | EX_ST(STORE_INIT(%o4, %o0 + 0x38)) | 270 | EX_ST(STORE_INIT(%o4, %o0 + 0x38)) |
216 | subcc %g1, 64, %g1 | 271 | subcc %g1, 64, %g1 |
217 | bne,pt %XCC, 1b | 272 | bne,pt %XCC, 1b |
218 | add %o0, 64, %o0 | 273 | add %o0, 64, %o0 |
219 | 274 | ||
220 | ba,pt %XCC, 60f | 275 | ba,pt %XCC, 60f |
221 | add %o1, 0x8, %o1 | 276 | add %i1, 0x8, %i1 |
222 | 277 | ||
223 | 50: /* Destination is 64-byte aligned, and source is 16-byte | 278 | 50: /* Destination is 64-byte aligned, and source is 16-byte |
224 | * aligned. | 279 | * aligned. |
225 | */ | 280 | */ |
226 | 1: EX_LD(LOAD_TWIN(%o1, %o4, %o5)) | 281 | mov 16, %o7 |
227 | add %o1, 16, %o1 | 282 | mov 32, %g2 |
228 | EX_LD(LOAD_TWIN(%o1, %g2, %g3)) | 283 | mov 48, %g3 |
229 | add %o1, 16 + 32, %o1 | 284 | mov 64, %o1 |
230 | LOAD(prefetch, %o1, #one_read) | 285 | 1: EX_LD(LOAD_TWIN(%i1 + %g0, %o4, %o5)) |
231 | sub %o1, 32, %o1 | 286 | EX_LD(LOAD_TWIN(%i1 + %o7, %o2, %o3)) |
287 | LOAD(prefetch, %i1 + %o1, #one_read) | ||
232 | EX_ST(STORE_INIT(%o4, %o0 + 0x00)) ! initializes cache line | 288 | EX_ST(STORE_INIT(%o4, %o0 + 0x00)) ! initializes cache line |
233 | EX_ST(STORE_INIT(%o5, %o0 + 0x08)) | 289 | EX_ST(STORE_INIT(%o5, %o0 + 0x08)) |
234 | EX_LD(LOAD_TWIN(%o1, %o4, %o5)) | 290 | EX_LD(LOAD_TWIN(%i1 + %g2, %o4, %o5)) |
235 | add %o1, 16, %o1 | 291 | EX_ST(STORE_INIT(%o2, %o0 + 0x10)) |
236 | EX_ST(STORE_INIT(%g2, %o0 + 0x10)) | 292 | EX_ST(STORE_INIT(%o3, %o0 + 0x18)) |
237 | EX_ST(STORE_INIT(%g3, %o0 + 0x18)) | 293 | EX_LD(LOAD_TWIN(%i1 + %g3, %o2, %o3)) |
238 | EX_LD(LOAD_TWIN(%o1, %g2, %g3)) | 294 | add %i1, 64, %i1 |
239 | add %o1, 16, %o1 | ||
240 | EX_ST(STORE_INIT(%o4, %o0 + 0x20)) | 295 | EX_ST(STORE_INIT(%o4, %o0 + 0x20)) |
241 | EX_ST(STORE_INIT(%o5, %o0 + 0x28)) | 296 | EX_ST(STORE_INIT(%o5, %o0 + 0x28)) |
242 | EX_ST(STORE_INIT(%g2, %o0 + 0x30)) | 297 | EX_ST(STORE_INIT(%o2, %o0 + 0x30)) |
243 | EX_ST(STORE_INIT(%g3, %o0 + 0x38)) | 298 | EX_ST(STORE_INIT(%o3, %o0 + 0x38)) |
244 | subcc %g1, 64, %g1 | 299 | subcc %g1, 64, %g1 |
245 | bne,pt %XCC, 1b | 300 | bne,pt %XCC, 1b |
246 | add %o0, 64, %o0 | 301 | add %o0, 64, %o0 |
@@ -249,47 +304,47 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ | |||
249 | 60: | 304 | 60: |
250 | membar #Sync | 305 | membar #Sync |
251 | 306 | ||
252 | /* %o2 contains any final bytes still needed to be copied | 307 | /* %i2 contains any final bytes still needed to be copied |
253 | * over. If anything is left, we copy it one byte at a time. | 308 | * over. If anything is left, we copy it one byte at a time. |
254 | */ | 309 | */ |
255 | RESTORE_ASI(%o3) | 310 | RESTORE_ASI(%i3) |
256 | brz,pt %o2, 85f | 311 | brz,pt %i2, 85f |
257 | sub %o0, %o1, %o3 | 312 | sub %o0, %i1, %i3 |
258 | ba,a,pt %XCC, 90f | 313 | ba,a,pt %XCC, 90f |
259 | 314 | ||
260 | .align 64 | 315 | .align 64 |
261 | 70: /* 16 < len <= 64 */ | 316 | 70: /* 16 < len <= 64 */ |
262 | bne,pn %XCC, 75f | 317 | bne,pn %XCC, 75f |
263 | sub %o0, %o1, %o3 | 318 | sub %o0, %i1, %i3 |
264 | 319 | ||
265 | 72: | 320 | 72: |
266 | andn %o2, 0xf, %o4 | 321 | andn %i2, 0xf, %i4 |
267 | and %o2, 0xf, %o2 | 322 | and %i2, 0xf, %i2 |
268 | 1: subcc %o4, 0x10, %o4 | 323 | 1: subcc %i4, 0x10, %i4 |
269 | EX_LD(LOAD(ldx, %o1, %o5)) | 324 | EX_LD(LOAD(ldx, %i1, %i5)) |
270 | add %o1, 0x08, %o1 | 325 | add %i1, 0x08, %i1 |
271 | EX_LD(LOAD(ldx, %o1, %g1)) | 326 | EX_LD(LOAD(ldx, %i1, %g1)) |
272 | sub %o1, 0x08, %o1 | 327 | sub %i1, 0x08, %i1 |
273 | EX_ST(STORE(stx, %o5, %o1 + %o3)) | 328 | EX_ST(STORE(stx, %i5, %i1 + %i3)) |
274 | add %o1, 0x8, %o1 | 329 | add %i1, 0x8, %i1 |
275 | EX_ST(STORE(stx, %g1, %o1 + %o3)) | 330 | EX_ST(STORE(stx, %g1, %i1 + %i3)) |
276 | bgu,pt %XCC, 1b | 331 | bgu,pt %XCC, 1b |
277 | add %o1, 0x8, %o1 | 332 | add %i1, 0x8, %i1 |
278 | 73: andcc %o2, 0x8, %g0 | 333 | 73: andcc %i2, 0x8, %g0 |
279 | be,pt %XCC, 1f | 334 | be,pt %XCC, 1f |
280 | nop | 335 | nop |
281 | sub %o2, 0x8, %o2 | 336 | sub %i2, 0x8, %i2 |
282 | EX_LD(LOAD(ldx, %o1, %o5)) | 337 | EX_LD(LOAD(ldx, %i1, %i5)) |
283 | EX_ST(STORE(stx, %o5, %o1 + %o3)) | 338 | EX_ST(STORE(stx, %i5, %i1 + %i3)) |
284 | add %o1, 0x8, %o1 | 339 | add %i1, 0x8, %i1 |
285 | 1: andcc %o2, 0x4, %g0 | 340 | 1: andcc %i2, 0x4, %g0 |
286 | be,pt %XCC, 1f | 341 | be,pt %XCC, 1f |
287 | nop | 342 | nop |
288 | sub %o2, 0x4, %o2 | 343 | sub %i2, 0x4, %i2 |
289 | EX_LD(LOAD(lduw, %o1, %o5)) | 344 | EX_LD(LOAD(lduw, %i1, %i5)) |
290 | EX_ST(STORE(stw, %o5, %o1 + %o3)) | 345 | EX_ST(STORE(stw, %i5, %i1 + %i3)) |
291 | add %o1, 0x4, %o1 | 346 | add %i1, 0x4, %i1 |
292 | 1: cmp %o2, 0 | 347 | 1: cmp %i2, 0 |
293 | be,pt %XCC, 85f | 348 | be,pt %XCC, 85f |
294 | nop | 349 | nop |
295 | ba,pt %xcc, 90f | 350 | ba,pt %xcc, 90f |
@@ -300,71 +355,71 @@ FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ | |||
300 | sub %g1, 0x8, %g1 | 355 | sub %g1, 0x8, %g1 |
301 | be,pn %icc, 2f | 356 | be,pn %icc, 2f |
302 | sub %g0, %g1, %g1 | 357 | sub %g0, %g1, %g1 |
303 | sub %o2, %g1, %o2 | 358 | sub %i2, %g1, %i2 |
304 | 359 | ||
305 | 1: subcc %g1, 1, %g1 | 360 | 1: subcc %g1, 1, %g1 |
306 | EX_LD(LOAD(ldub, %o1, %o5)) | 361 | EX_LD(LOAD(ldub, %i1, %i5)) |
307 | EX_ST(STORE(stb, %o5, %o1 + %o3)) | 362 | EX_ST(STORE(stb, %i5, %i1 + %i3)) |
308 | bgu,pt %icc, 1b | 363 | bgu,pt %icc, 1b |
309 | add %o1, 1, %o1 | 364 | add %i1, 1, %i1 |
310 | 365 | ||
311 | 2: add %o1, %o3, %o0 | 366 | 2: add %i1, %i3, %o0 |
312 | andcc %o1, 0x7, %g1 | 367 | andcc %i1, 0x7, %g1 |
313 | bne,pt %icc, 8f | 368 | bne,pt %icc, 8f |
314 | sll %g1, 3, %g1 | 369 | sll %g1, 3, %g1 |
315 | 370 | ||
316 | cmp %o2, 16 | 371 | cmp %i2, 16 |
317 | bgeu,pt %icc, 72b | 372 | bgeu,pt %icc, 72b |
318 | nop | 373 | nop |
319 | ba,a,pt %xcc, 73b | 374 | ba,a,pt %xcc, 73b |
320 | 375 | ||
321 | 8: mov 64, %o3 | 376 | 8: mov 64, %i3 |
322 | andn %o1, 0x7, %o1 | 377 | andn %i1, 0x7, %i1 |
323 | EX_LD(LOAD(ldx, %o1, %g2)) | 378 | EX_LD(LOAD(ldx, %i1, %g2)) |
324 | sub %o3, %g1, %o3 | 379 | sub %i3, %g1, %i3 |
325 | andn %o2, 0x7, %o4 | 380 | andn %i2, 0x7, %i4 |
326 | sllx %g2, %g1, %g2 | 381 | sllx %g2, %g1, %g2 |
327 | 1: add %o1, 0x8, %o1 | 382 | 1: add %i1, 0x8, %i1 |
328 | EX_LD(LOAD(ldx, %o1, %g3)) | 383 | EX_LD(LOAD(ldx, %i1, %g3)) |
329 | subcc %o4, 0x8, %o4 | 384 | subcc %i4, 0x8, %i4 |
330 | srlx %g3, %o3, %o5 | 385 | srlx %g3, %i3, %i5 |
331 | or %o5, %g2, %o5 | 386 | or %i5, %g2, %i5 |
332 | EX_ST(STORE(stx, %o5, %o0)) | 387 | EX_ST(STORE(stx, %i5, %o0)) |
333 | add %o0, 0x8, %o0 | 388 | add %o0, 0x8, %o0 |
334 | bgu,pt %icc, 1b | 389 | bgu,pt %icc, 1b |
335 | sllx %g3, %g1, %g2 | 390 | sllx %g3, %g1, %g2 |
336 | 391 | ||
337 | srl %g1, 3, %g1 | 392 | srl %g1, 3, %g1 |
338 | andcc %o2, 0x7, %o2 | 393 | andcc %i2, 0x7, %i2 |
339 | be,pn %icc, 85f | 394 | be,pn %icc, 85f |
340 | add %o1, %g1, %o1 | 395 | add %i1, %g1, %i1 |
341 | ba,pt %xcc, 90f | 396 | ba,pt %xcc, 90f |
342 | sub %o0, %o1, %o3 | 397 | sub %o0, %i1, %i3 |
343 | 398 | ||
344 | .align 64 | 399 | .align 64 |
345 | 80: /* 0 < len <= 16 */ | 400 | 80: /* 0 < len <= 16 */ |
346 | andcc %o3, 0x3, %g0 | 401 | andcc %i3, 0x3, %g0 |
347 | bne,pn %XCC, 90f | 402 | bne,pn %XCC, 90f |
348 | sub %o0, %o1, %o3 | 403 | sub %o0, %i1, %i3 |
349 | 404 | ||
350 | 1: | 405 | 1: |
351 | subcc %o2, 4, %o2 | 406 | subcc %i2, 4, %i2 |
352 | EX_LD(LOAD(lduw, %o1, %g1)) | 407 | EX_LD(LOAD(lduw, %i1, %g1)) |
353 | EX_ST(STORE(stw, %g1, %o1 + %o3)) | 408 | EX_ST(STORE(stw, %g1, %i1 + %i3)) |
354 | bgu,pt %XCC, 1b | 409 | bgu,pt %XCC, 1b |
355 | add %o1, 4, %o1 | 410 | add %i1, 4, %i1 |
356 | 411 | ||
357 | 85: retl | 412 | 85: ret |
358 | mov EX_RETVAL(GLOBAL_SPARE), %o0 | 413 | restore EX_RETVAL(%i0), %g0, %o0 |
359 | 414 | ||
360 | .align 32 | 415 | .align 32 |
361 | 90: | 416 | 90: |
362 | subcc %o2, 1, %o2 | 417 | subcc %i2, 1, %i2 |
363 | EX_LD(LOAD(ldub, %o1, %g1)) | 418 | EX_LD(LOAD(ldub, %i1, %g1)) |
364 | EX_ST(STORE(stb, %g1, %o1 + %o3)) | 419 | EX_ST(STORE(stb, %g1, %i1 + %i3)) |
365 | bgu,pt %XCC, 90b | 420 | bgu,pt %XCC, 90b |
366 | add %o1, 1, %o1 | 421 | add %i1, 1, %i1 |
367 | retl | 422 | ret |
368 | mov EX_RETVAL(GLOBAL_SPARE), %o0 | 423 | restore EX_RETVAL(%i0), %g0, %o0 |
369 | 424 | ||
370 | .size FUNC_NAME, .-FUNC_NAME | 425 | .size FUNC_NAME, .-FUNC_NAME |