Henrik Smiding | 86a1600 | 2014-05-16 13:26:08 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2014 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
Henrik Smiding | 86a1600 | 2014-05-16 13:26:08 +0200 | [diff] [blame] | 16 | |
| 17 | #include "cache.h" |
| 18 | |
Varvara Rainchik | 458d125 | 2014-09-08 16:27:01 +0400 | [diff] [blame^] | 19 | #ifndef MEMSET |
| 20 | # define MEMSET android_memset32 |
| 21 | #endif |
| 22 | |
Henrik Smiding | 86a1600 | 2014-05-16 13:26:08 +0200 | [diff] [blame] | 23 | #ifndef L |
| 24 | # define L(label) .L##label |
| 25 | #endif |
| 26 | |
| 27 | #ifndef ALIGN |
| 28 | # define ALIGN(n) .p2align n |
| 29 | #endif |
| 30 | |
| 31 | #ifndef cfi_startproc |
| 32 | # define cfi_startproc .cfi_startproc |
| 33 | #endif |
| 34 | |
| 35 | #ifndef cfi_endproc |
| 36 | # define cfi_endproc .cfi_endproc |
| 37 | #endif |
| 38 | |
| 39 | #ifndef ENTRY |
| 40 | # define ENTRY(name) \ |
| 41 | .type name, @function; \ |
| 42 | .globl name; \ |
| 43 | .p2align 4; \ |
| 44 | name: \ |
| 45 | cfi_startproc |
| 46 | #endif |
| 47 | |
| 48 | #ifndef END |
| 49 | # define END(name) \ |
| 50 | cfi_endproc; \ |
| 51 | .size name, .-name |
| 52 | #endif |
| 53 | |
| 54 | #define JMPTBL(I, B) I - B |
| 55 | |
| 56 | /* Branch to an entry in a jump table. TABLE is a jump table with |
| 57 | relative offsets. INDEX is a register contains the index into the |
| 58 | jump table. SCALE is the scale of INDEX. */ |
| 59 | #define BRANCH_TO_JMPTBL_ENTRY(TABLE, INDEX, SCALE) \ |
| 60 | lea TABLE(%rip), %r11; \ |
| 61 | movslq (%r11, INDEX, SCALE), INDEX; \ |
| 62 | lea (%r11, INDEX), INDEX; \ |
| 63 | jmp *INDEX |
| 64 | |
| 65 | .section .text.sse2,"ax",@progbits |
| 66 | ALIGN (4) |
Varvara Rainchik | 458d125 | 2014-09-08 16:27:01 +0400 | [diff] [blame^] | 67 | ENTRY (MEMSET) // Address in rdi |
Henrik Smiding | 86a1600 | 2014-05-16 13:26:08 +0200 | [diff] [blame] | 68 | shr $2, %rdx // Count in rdx |
| 69 | movl %esi, %ecx // Pattern in ecx |
| 70 | |
| 71 | cmp $16, %rdx |
| 72 | jae L(16dbwordsormore) |
| 73 | |
| 74 | L(write_less16dbwords): |
| 75 | lea (%rdi, %rdx, 4), %rdi |
| 76 | BRANCH_TO_JMPTBL_ENTRY (L(table_less16dbwords), %rdx, 4) |
| 77 | |
| 78 | .pushsection .rodata.sse2,"a",@progbits |
| 79 | ALIGN (2) |
| 80 | L(table_less16dbwords): |
| 81 | .int JMPTBL (L(write_0dbwords), L(table_less16dbwords)) |
| 82 | .int JMPTBL (L(write_1dbwords), L(table_less16dbwords)) |
| 83 | .int JMPTBL (L(write_2dbwords), L(table_less16dbwords)) |
| 84 | .int JMPTBL (L(write_3dbwords), L(table_less16dbwords)) |
| 85 | .int JMPTBL (L(write_4dbwords), L(table_less16dbwords)) |
| 86 | .int JMPTBL (L(write_5dbwords), L(table_less16dbwords)) |
| 87 | .int JMPTBL (L(write_6dbwords), L(table_less16dbwords)) |
| 88 | .int JMPTBL (L(write_7dbwords), L(table_less16dbwords)) |
| 89 | .int JMPTBL (L(write_8dbwords), L(table_less16dbwords)) |
| 90 | .int JMPTBL (L(write_9dbwords), L(table_less16dbwords)) |
| 91 | .int JMPTBL (L(write_10dbwords), L(table_less16dbwords)) |
| 92 | .int JMPTBL (L(write_11dbwords), L(table_less16dbwords)) |
| 93 | .int JMPTBL (L(write_12dbwords), L(table_less16dbwords)) |
| 94 | .int JMPTBL (L(write_13dbwords), L(table_less16dbwords)) |
| 95 | .int JMPTBL (L(write_14dbwords), L(table_less16dbwords)) |
| 96 | .int JMPTBL (L(write_15dbwords), L(table_less16dbwords)) |
| 97 | .popsection |
| 98 | |
| 99 | ALIGN (4) |
| 100 | L(write_15dbwords): |
| 101 | movl %ecx, -60(%rdi) |
| 102 | L(write_14dbwords): |
| 103 | movl %ecx, -56(%rdi) |
| 104 | L(write_13dbwords): |
| 105 | movl %ecx, -52(%rdi) |
| 106 | L(write_12dbwords): |
| 107 | movl %ecx, -48(%rdi) |
| 108 | L(write_11dbwords): |
| 109 | movl %ecx, -44(%rdi) |
| 110 | L(write_10dbwords): |
| 111 | movl %ecx, -40(%rdi) |
| 112 | L(write_9dbwords): |
| 113 | movl %ecx, -36(%rdi) |
| 114 | L(write_8dbwords): |
| 115 | movl %ecx, -32(%rdi) |
| 116 | L(write_7dbwords): |
| 117 | movl %ecx, -28(%rdi) |
| 118 | L(write_6dbwords): |
| 119 | movl %ecx, -24(%rdi) |
| 120 | L(write_5dbwords): |
| 121 | movl %ecx, -20(%rdi) |
| 122 | L(write_4dbwords): |
| 123 | movl %ecx, -16(%rdi) |
| 124 | L(write_3dbwords): |
| 125 | movl %ecx, -12(%rdi) |
| 126 | L(write_2dbwords): |
| 127 | movl %ecx, -8(%rdi) |
| 128 | L(write_1dbwords): |
| 129 | movl %ecx, -4(%rdi) |
| 130 | L(write_0dbwords): |
| 131 | ret |
| 132 | |
| 133 | ALIGN (4) |
| 134 | L(16dbwordsormore): |
| 135 | test $3, %edi |
| 136 | jz L(aligned4bytes) |
| 137 | mov %ecx, (%rdi) |
| 138 | mov %ecx, -4(%rdi, %rdx, 4) |
| 139 | sub $1, %rdx |
| 140 | rol $24, %ecx |
| 141 | add $1, %rdi |
| 142 | test $3, %edi |
| 143 | jz L(aligned4bytes) |
| 144 | ror $8, %ecx |
| 145 | add $1, %rdi |
| 146 | test $3, %edi |
| 147 | jz L(aligned4bytes) |
| 148 | ror $8, %ecx |
| 149 | add $1, %rdi |
| 150 | L(aligned4bytes): |
| 151 | shl $2, %rdx |
| 152 | |
| 153 | /* Fill xmm0 with the pattern. */ |
| 154 | movd %ecx, %xmm0 |
| 155 | pshufd $0, %xmm0, %xmm0 |
| 156 | |
| 157 | testl $0xf, %edi |
| 158 | jz L(aligned_16) |
| 159 | /* RDX > 32 and RDI is not 16 byte aligned. */ |
| 160 | movdqu %xmm0, (%rdi) |
| 161 | mov %rdi, %rsi |
| 162 | and $-16, %rdi |
| 163 | add $16, %rdi |
| 164 | sub %rdi, %rsi |
| 165 | add %rsi, %rdx |
| 166 | |
| 167 | ALIGN (4) |
| 168 | L(aligned_16): |
| 169 | cmp $128, %rdx |
| 170 | jge L(128bytesormore) |
| 171 | |
| 172 | L(aligned_16_less128bytes): |
| 173 | add %rdx, %rdi |
| 174 | shr $2, %rdx |
| 175 | BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes), %rdx, 4) |
| 176 | |
| 177 | ALIGN (4) |
| 178 | L(128bytesormore): |
| 179 | cmp $SHARED_CACHE_SIZE, %rdx |
| 180 | jg L(128bytesormore_nt) |
| 181 | |
| 182 | L(128bytesormore_normal): |
| 183 | sub $128, %rdx |
| 184 | movdqa %xmm0, (%rdi) |
| 185 | movdqa %xmm0, 0x10(%rdi) |
| 186 | movdqa %xmm0, 0x20(%rdi) |
| 187 | movdqa %xmm0, 0x30(%rdi) |
| 188 | movdqa %xmm0, 0x40(%rdi) |
| 189 | movdqa %xmm0, 0x50(%rdi) |
| 190 | movdqa %xmm0, 0x60(%rdi) |
| 191 | movdqa %xmm0, 0x70(%rdi) |
| 192 | lea 128(%rdi), %rdi |
| 193 | cmp $128, %rdx |
| 194 | jl L(128bytesless_normal) |
| 195 | |
| 196 | sub $128, %rdx |
| 197 | movdqa %xmm0, (%rdi) |
| 198 | movdqa %xmm0, 0x10(%rdi) |
| 199 | movdqa %xmm0, 0x20(%rdi) |
| 200 | movdqa %xmm0, 0x30(%rdi) |
| 201 | movdqa %xmm0, 0x40(%rdi) |
| 202 | movdqa %xmm0, 0x50(%rdi) |
| 203 | movdqa %xmm0, 0x60(%rdi) |
| 204 | movdqa %xmm0, 0x70(%rdi) |
| 205 | lea 128(%rdi), %rdi |
| 206 | cmp $128, %rdx |
| 207 | jl L(128bytesless_normal) |
| 208 | |
| 209 | sub $128, %rdx |
| 210 | movdqa %xmm0, (%rdi) |
| 211 | movdqa %xmm0, 0x10(%rdi) |
| 212 | movdqa %xmm0, 0x20(%rdi) |
| 213 | movdqa %xmm0, 0x30(%rdi) |
| 214 | movdqa %xmm0, 0x40(%rdi) |
| 215 | movdqa %xmm0, 0x50(%rdi) |
| 216 | movdqa %xmm0, 0x60(%rdi) |
| 217 | movdqa %xmm0, 0x70(%rdi) |
| 218 | lea 128(%rdi), %rdi |
| 219 | cmp $128, %rdx |
| 220 | jl L(128bytesless_normal) |
| 221 | |
| 222 | sub $128, %rdx |
| 223 | movdqa %xmm0, (%rdi) |
| 224 | movdqa %xmm0, 0x10(%rdi) |
| 225 | movdqa %xmm0, 0x20(%rdi) |
| 226 | movdqa %xmm0, 0x30(%rdi) |
| 227 | movdqa %xmm0, 0x40(%rdi) |
| 228 | movdqa %xmm0, 0x50(%rdi) |
| 229 | movdqa %xmm0, 0x60(%rdi) |
| 230 | movdqa %xmm0, 0x70(%rdi) |
| 231 | lea 128(%rdi), %rdi |
| 232 | cmp $128, %rdx |
| 233 | jge L(128bytesormore_normal) |
| 234 | |
| 235 | L(128bytesless_normal): |
| 236 | add %rdx, %rdi |
| 237 | shr $2, %rdx |
| 238 | BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes), %rdx, 4) |
| 239 | |
| 240 | ALIGN (4) |
| 241 | L(128bytesormore_nt): |
| 242 | sub $128, %rdx |
| 243 | movntdq %xmm0, (%rdi) |
| 244 | movntdq %xmm0, 0x10(%rdi) |
| 245 | movntdq %xmm0, 0x20(%rdi) |
| 246 | movntdq %xmm0, 0x30(%rdi) |
| 247 | movntdq %xmm0, 0x40(%rdi) |
| 248 | movntdq %xmm0, 0x50(%rdi) |
| 249 | movntdq %xmm0, 0x60(%rdi) |
| 250 | movntdq %xmm0, 0x70(%rdi) |
| 251 | lea 128(%rdi), %rdi |
| 252 | cmp $128, %rdx |
| 253 | jge L(128bytesormore_nt) |
| 254 | |
| 255 | sfence |
| 256 | add %rdx, %rdi |
| 257 | shr $2, %rdx |
| 258 | BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes), %rdx, 4) |
| 259 | |
| 260 | .pushsection .rodata.sse2,"a",@progbits |
| 261 | ALIGN (2) |
| 262 | L(table_16_128bytes): |
| 263 | .int JMPTBL (L(aligned_16_0bytes), L(table_16_128bytes)) |
| 264 | .int JMPTBL (L(aligned_16_4bytes), L(table_16_128bytes)) |
| 265 | .int JMPTBL (L(aligned_16_8bytes), L(table_16_128bytes)) |
| 266 | .int JMPTBL (L(aligned_16_12bytes), L(table_16_128bytes)) |
| 267 | .int JMPTBL (L(aligned_16_16bytes), L(table_16_128bytes)) |
| 268 | .int JMPTBL (L(aligned_16_20bytes), L(table_16_128bytes)) |
| 269 | .int JMPTBL (L(aligned_16_24bytes), L(table_16_128bytes)) |
| 270 | .int JMPTBL (L(aligned_16_28bytes), L(table_16_128bytes)) |
| 271 | .int JMPTBL (L(aligned_16_32bytes), L(table_16_128bytes)) |
| 272 | .int JMPTBL (L(aligned_16_36bytes), L(table_16_128bytes)) |
| 273 | .int JMPTBL (L(aligned_16_40bytes), L(table_16_128bytes)) |
| 274 | .int JMPTBL (L(aligned_16_44bytes), L(table_16_128bytes)) |
| 275 | .int JMPTBL (L(aligned_16_48bytes), L(table_16_128bytes)) |
| 276 | .int JMPTBL (L(aligned_16_52bytes), L(table_16_128bytes)) |
| 277 | .int JMPTBL (L(aligned_16_56bytes), L(table_16_128bytes)) |
| 278 | .int JMPTBL (L(aligned_16_60bytes), L(table_16_128bytes)) |
| 279 | .int JMPTBL (L(aligned_16_64bytes), L(table_16_128bytes)) |
| 280 | .int JMPTBL (L(aligned_16_68bytes), L(table_16_128bytes)) |
| 281 | .int JMPTBL (L(aligned_16_72bytes), L(table_16_128bytes)) |
| 282 | .int JMPTBL (L(aligned_16_76bytes), L(table_16_128bytes)) |
| 283 | .int JMPTBL (L(aligned_16_80bytes), L(table_16_128bytes)) |
| 284 | .int JMPTBL (L(aligned_16_84bytes), L(table_16_128bytes)) |
| 285 | .int JMPTBL (L(aligned_16_88bytes), L(table_16_128bytes)) |
| 286 | .int JMPTBL (L(aligned_16_92bytes), L(table_16_128bytes)) |
| 287 | .int JMPTBL (L(aligned_16_96bytes), L(table_16_128bytes)) |
| 288 | .int JMPTBL (L(aligned_16_100bytes), L(table_16_128bytes)) |
| 289 | .int JMPTBL (L(aligned_16_104bytes), L(table_16_128bytes)) |
| 290 | .int JMPTBL (L(aligned_16_108bytes), L(table_16_128bytes)) |
| 291 | .int JMPTBL (L(aligned_16_112bytes), L(table_16_128bytes)) |
| 292 | .int JMPTBL (L(aligned_16_116bytes), L(table_16_128bytes)) |
| 293 | .int JMPTBL (L(aligned_16_120bytes), L(table_16_128bytes)) |
| 294 | .int JMPTBL (L(aligned_16_124bytes), L(table_16_128bytes)) |
| 295 | .popsection |
| 296 | |
| 297 | ALIGN (4) |
| 298 | L(aligned_16_112bytes): |
| 299 | movdqa %xmm0, -112(%rdi) |
| 300 | L(aligned_16_96bytes): |
| 301 | movdqa %xmm0, -96(%rdi) |
| 302 | L(aligned_16_80bytes): |
| 303 | movdqa %xmm0, -80(%rdi) |
| 304 | L(aligned_16_64bytes): |
| 305 | movdqa %xmm0, -64(%rdi) |
| 306 | L(aligned_16_48bytes): |
| 307 | movdqa %xmm0, -48(%rdi) |
| 308 | L(aligned_16_32bytes): |
| 309 | movdqa %xmm0, -32(%rdi) |
| 310 | L(aligned_16_16bytes): |
| 311 | movdqa %xmm0, -16(%rdi) |
| 312 | L(aligned_16_0bytes): |
| 313 | ret |
| 314 | |
| 315 | ALIGN (4) |
| 316 | L(aligned_16_116bytes): |
| 317 | movdqa %xmm0, -116(%rdi) |
| 318 | L(aligned_16_100bytes): |
| 319 | movdqa %xmm0, -100(%rdi) |
| 320 | L(aligned_16_84bytes): |
| 321 | movdqa %xmm0, -84(%rdi) |
| 322 | L(aligned_16_68bytes): |
| 323 | movdqa %xmm0, -68(%rdi) |
| 324 | L(aligned_16_52bytes): |
| 325 | movdqa %xmm0, -52(%rdi) |
| 326 | L(aligned_16_36bytes): |
| 327 | movdqa %xmm0, -36(%rdi) |
| 328 | L(aligned_16_20bytes): |
| 329 | movdqa %xmm0, -20(%rdi) |
| 330 | L(aligned_16_4bytes): |
| 331 | movl %ecx, -4(%rdi) |
| 332 | ret |
| 333 | |
| 334 | ALIGN (4) |
| 335 | L(aligned_16_120bytes): |
| 336 | movdqa %xmm0, -120(%rdi) |
| 337 | L(aligned_16_104bytes): |
| 338 | movdqa %xmm0, -104(%rdi) |
| 339 | L(aligned_16_88bytes): |
| 340 | movdqa %xmm0, -88(%rdi) |
| 341 | L(aligned_16_72bytes): |
| 342 | movdqa %xmm0, -72(%rdi) |
| 343 | L(aligned_16_56bytes): |
| 344 | movdqa %xmm0, -56(%rdi) |
| 345 | L(aligned_16_40bytes): |
| 346 | movdqa %xmm0, -40(%rdi) |
| 347 | L(aligned_16_24bytes): |
| 348 | movdqa %xmm0, -24(%rdi) |
| 349 | L(aligned_16_8bytes): |
| 350 | movq %xmm0, -8(%rdi) |
| 351 | ret |
| 352 | |
| 353 | ALIGN (4) |
| 354 | L(aligned_16_124bytes): |
| 355 | movdqa %xmm0, -124(%rdi) |
| 356 | L(aligned_16_108bytes): |
| 357 | movdqa %xmm0, -108(%rdi) |
| 358 | L(aligned_16_92bytes): |
| 359 | movdqa %xmm0, -92(%rdi) |
| 360 | L(aligned_16_76bytes): |
| 361 | movdqa %xmm0, -76(%rdi) |
| 362 | L(aligned_16_60bytes): |
| 363 | movdqa %xmm0, -60(%rdi) |
| 364 | L(aligned_16_44bytes): |
| 365 | movdqa %xmm0, -44(%rdi) |
| 366 | L(aligned_16_28bytes): |
| 367 | movdqa %xmm0, -28(%rdi) |
| 368 | L(aligned_16_12bytes): |
| 369 | movq %xmm0, -12(%rdi) |
| 370 | movl %ecx, -4(%rdi) |
| 371 | ret |
| 372 | |
Varvara Rainchik | 458d125 | 2014-09-08 16:27:01 +0400 | [diff] [blame^] | 373 | END (MEMSET) |