1/* NG4memcpy.S: Niagara-4 optimized memcpy. 2 * 3 * Copyright (C) 2012 David S. Miller (davem@davemloft.net) 4 */ 5 6#ifdef __KERNEL__ 7#include <asm/visasm.h> 8#include <asm/asi.h> 9#define GLOBAL_SPARE %g7 10#else 11#define ASI_BLK_INIT_QUAD_LDD_P 0xe2 12#define FPRS_FEF 0x04 13 14/* On T4 it is very expensive to access ASRs like %fprs and 15 * %asi, avoiding a read or a write can save ~50 cycles. 16 */ 17#define FPU_ENTER \ 18 rd %fprs, %o5; \ 19 andcc %o5, FPRS_FEF, %g0; \ 20 be,a,pn %icc, 999f; \ 21 wr %g0, FPRS_FEF, %fprs; \ 22 999: 23 24#ifdef MEMCPY_DEBUG 25#define VISEntryHalf FPU_ENTER; \ 26 clr %g1; clr %g2; clr %g3; clr %g5; subcc %g0, %g0, %g0; 27#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs 28#else 29#define VISEntryHalf FPU_ENTER 30#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs 31#endif 32 33#define GLOBAL_SPARE %g5 34#endif 35 36#ifndef STORE_ASI 37#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA 38#define STORE_ASI ASI_BLK_INIT_QUAD_LDD_P 39#else 40#define STORE_ASI 0x80 /* ASI_P */ 41#endif 42#endif 43 44#ifndef EX_LD 45#define EX_LD(x) x 46#endif 47 48#ifndef EX_ST 49#define EX_ST(x) x 50#endif 51 52#ifndef EX_RETVAL 53#define EX_RETVAL(x) x 54#endif 55 56#ifndef LOAD 57#define LOAD(type,addr,dest) type [addr], dest 58#endif 59 60#ifndef STORE 61#ifndef MEMCPY_DEBUG 62#define STORE(type,src,addr) type src, [addr] 63#else 64#define STORE(type,src,addr) type##a src, [addr] %asi 65#endif 66#endif 67 68#ifndef STORE_INIT 69#define STORE_INIT(src,addr) stxa src, [addr] STORE_ASI 70#endif 71 72#ifndef FUNC_NAME 73#define FUNC_NAME NG4memcpy 74#endif 75#ifndef PREAMBLE 76#define PREAMBLE 77#endif 78 79#ifndef XCC 80#define XCC xcc 81#endif 82 83 .register %g2,#scratch 84 .register %g3,#scratch 85 86 .text 87 .align 64 88 89 .globl FUNC_NAME 90 .type FUNC_NAME,#function 91FUNC_NAME: /* %o0=dst, %o1=src, %o2=len */ 92#ifdef MEMCPY_DEBUG 93 wr %g0, 0x80, %asi 94#endif 95 srlx %o2, 31, %g2 96 cmp %g2, 0 97 tne %XCC, 5 98 PREAMBLE 99 mov %o0, %o3 100 brz,pn %o2, .Lexit 101 cmp %o2, 3 102 ble,pn %icc, .Ltiny 103 cmp %o2, 19 104 ble,pn %icc, .Lsmall 105 or %o0, %o1, %g2 106 cmp %o2, 128 107 bl,pn %icc, .Lmedium 108 nop 109 110.Llarge:/* len >= 0x80 */ 111 /* First get dest 8 byte aligned. */ 112 sub %g0, %o0, %g1 113 and %g1, 0x7, %g1 114 brz,pt %g1, 51f 115 sub %o2, %g1, %o2 116 1171: EX_LD(LOAD(ldub, %o1 + 0x00, %g2)) 118 add %o1, 1, %o1 119 subcc %g1, 1, %g1 120 add %o0, 1, %o0 121 bne,pt %icc, 1b 122 EX_ST(STORE(stb, %g2, %o0 - 0x01)) 123 12451: LOAD(prefetch, %o1 + 0x040, #n_reads_strong) 125 LOAD(prefetch, %o1 + 0x080, #n_reads_strong) 126 LOAD(prefetch, %o1 + 0x0c0, #n_reads_strong) 127 LOAD(prefetch, %o1 + 0x100, #n_reads_strong) 128 LOAD(prefetch, %o1 + 0x140, #n_reads_strong) 129 LOAD(prefetch, %o1 + 0x180, #n_reads_strong) 130 LOAD(prefetch, %o1 + 0x1c0, #n_reads_strong) 131 LOAD(prefetch, %o1 + 0x200, #n_reads_strong) 132 133 /* Check if we can use the straight fully aligned 134 * loop, or we require the alignaddr/faligndata variant. 135 */ 136 andcc %o1, 0x7, %o5 137 bne,pn %icc, .Llarge_src_unaligned 138 sub %g0, %o0, %g1 139 140 /* Legitimize the use of initializing stores by getting dest 141 * to be 64-byte aligned. 142 */ 143 and %g1, 0x3f, %g1 144 brz,pt %g1, .Llarge_aligned 145 sub %o2, %g1, %o2 146 1471: EX_LD(LOAD(ldx, %o1 + 0x00, %g2)) 148 add %o1, 8, %o1 149 subcc %g1, 8, %g1 150 add %o0, 8, %o0 151 bne,pt %icc, 1b 152 EX_ST(STORE(stx, %g2, %o0 - 0x08)) 153 154.Llarge_aligned: 155 /* len >= 0x80 && src 8-byte aligned && dest 8-byte aligned */ 156 andn %o2, 0x3f, %o4 157 sub %o2, %o4, %o2 158 1591: EX_LD(LOAD(ldx, %o1 + 0x00, %g1)) 160 add %o1, 0x40, %o1 161 EX_LD(LOAD(ldx, %o1 - 0x38, %g2)) 162 subcc %o4, 0x40, %o4 163 EX_LD(LOAD(ldx, %o1 - 0x30, %g3)) 164 EX_LD(LOAD(ldx, %o1 - 0x28, GLOBAL_SPARE)) 165 EX_LD(LOAD(ldx, %o1 - 0x20, %o5)) 166 EX_ST(STORE_INIT(%g1, %o0)) 167 add %o0, 0x08, %o0 168 EX_ST(STORE_INIT(%g2, %o0)) 169 add %o0, 0x08, %o0 170 EX_LD(LOAD(ldx, %o1 - 0x18, %g2)) 171 EX_ST(STORE_INIT(%g3, %o0)) 172 add %o0, 0x08, %o0 173 EX_LD(LOAD(ldx, %o1 - 0x10, %g3)) 174 EX_ST(STORE_INIT(GLOBAL_SPARE, %o0)) 175 add %o0, 0x08, %o0 176 EX_LD(LOAD(ldx, %o1 - 0x08, GLOBAL_SPARE)) 177 EX_ST(STORE_INIT(%o5, %o0)) 178 add %o0, 0x08, %o0 179 EX_ST(STORE_INIT(%g2, %o0)) 180 add %o0, 0x08, %o0 181 EX_ST(STORE_INIT(%g3, %o0)) 182 add %o0, 0x08, %o0 183 EX_ST(STORE_INIT(GLOBAL_SPARE, %o0)) 184 add %o0, 0x08, %o0 185 bne,pt %icc, 1b 186 LOAD(prefetch, %o1 + 0x200, #n_reads_strong) 187 188 membar #StoreLoad | #StoreStore 189 190 brz,pn %o2, .Lexit 191 cmp %o2, 19 192 ble,pn %icc, .Lsmall_unaligned 193 nop 194 ba,a,pt %icc, .Lmedium_noprefetch 195 196.Lexit: retl 197 mov EX_RETVAL(%o3), %o0 198 199.Llarge_src_unaligned: 200 andn %o2, 0x3f, %o4 201 sub %o2, %o4, %o2 202 VISEntryHalf 203 alignaddr %o1, %g0, %g1 204 add %o1, %o4, %o1 205 EX_LD(LOAD(ldd, %g1 + 0x00, %f0)) 2061: EX_LD(LOAD(ldd, %g1 + 0x08, %f2)) 207 subcc %o4, 0x40, %o4 208 EX_LD(LOAD(ldd, %g1 + 0x10, %f4)) 209 EX_LD(LOAD(ldd, %g1 + 0x18, %f6)) 210 EX_LD(LOAD(ldd, %g1 + 0x20, %f8)) 211 EX_LD(LOAD(ldd, %g1 + 0x28, %f10)) 212 EX_LD(LOAD(ldd, %g1 + 0x30, %f12)) 213 EX_LD(LOAD(ldd, %g1 + 0x38, %f14)) 214 faligndata %f0, %f2, %f16 215 EX_LD(LOAD(ldd, %g1 + 0x40, %f0)) 216 faligndata %f2, %f4, %f18 217 add %g1, 0x40, %g1 218 faligndata %f4, %f6, %f20 219 faligndata %f6, %f8, %f22 220 faligndata %f8, %f10, %f24 221 faligndata %f10, %f12, %f26 222 faligndata %f12, %f14, %f28 223 faligndata %f14, %f0, %f30 224 EX_ST(STORE(std, %f16, %o0 + 0x00)) 225 EX_ST(STORE(std, %f18, %o0 + 0x08)) 226 EX_ST(STORE(std, %f20, %o0 + 0x10)) 227 EX_ST(STORE(std, %f22, %o0 + 0x18)) 228 EX_ST(STORE(std, %f24, %o0 + 0x20)) 229 EX_ST(STORE(std, %f26, %o0 + 0x28)) 230 EX_ST(STORE(std, %f28, %o0 + 0x30)) 231 EX_ST(STORE(std, %f30, %o0 + 0x38)) 232 add %o0, 0x40, %o0 233 bne,pt %icc, 1b 234 LOAD(prefetch, %g1 + 0x200, #n_reads_strong) 235 VISExitHalf 236 237 brz,pn %o2, .Lexit 238 cmp %o2, 19 239 ble,pn %icc, .Lsmall_unaligned 240 nop 241 ba,a,pt %icc, .Lmedium_unaligned 242 243.Lmedium: 244 LOAD(prefetch, %o1 + 0x40, #n_reads_strong) 245 andcc %g2, 0x7, %g0 246 bne,pn %icc, .Lmedium_unaligned 247 nop 248.Lmedium_noprefetch: 249 andncc %o2, 0x20 - 1, %o5 250 be,pn %icc, 2f 251 sub %o2, %o5, %o2 2521: EX_LD(LOAD(ldx, %o1 + 0x00, %g1)) 253 EX_LD(LOAD(ldx, %o1 + 0x08, %g2)) 254 EX_LD(LOAD(ldx, %o1 + 0x10, GLOBAL_SPARE)) 255 EX_LD(LOAD(ldx, %o1 + 0x18, %o4)) 256 add %o1, 0x20, %o1 257 subcc %o5, 0x20, %o5 258 EX_ST(STORE(stx, %g1, %o0 + 0x00)) 259 EX_ST(STORE(stx, %g2, %o0 + 0x08)) 260 EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x10)) 261 EX_ST(STORE(stx, %o4, %o0 + 0x18)) 262 bne,pt %icc, 1b 263 add %o0, 0x20, %o0 2642: andcc %o2, 0x18, %o5 265 be,pt %icc, 3f 266 sub %o2, %o5, %o2 2671: EX_LD(LOAD(ldx, %o1 + 0x00, %g1)) 268 add %o1, 0x08, %o1 269 add %o0, 0x08, %o0 270 subcc %o5, 0x08, %o5 271 bne,pt %icc, 1b 272 EX_ST(STORE(stx, %g1, %o0 - 0x08)) 2733: brz,pt %o2, .Lexit 274 cmp %o2, 0x04 275 bl,pn %icc, .Ltiny 276 nop 277 EX_LD(LOAD(lduw, %o1 + 0x00, %g1)) 278 add %o1, 0x04, %o1 279 add %o0, 0x04, %o0 280 subcc %o2, 0x04, %o2 281 bne,pn %icc, .Ltiny 282 EX_ST(STORE(stw, %g1, %o0 - 0x04)) 283 ba,a,pt %icc, .Lexit 284.Lmedium_unaligned: 285 /* First get dest 8 byte aligned. */ 286 sub %g0, %o0, %g1 287 and %g1, 0x7, %g1 288 brz,pt %g1, 2f 289 sub %o2, %g1, %o2 290 2911: EX_LD(LOAD(ldub, %o1 + 0x00, %g2)) 292 add %o1, 1, %o1 293 subcc %g1, 1, %g1 294 add %o0, 1, %o0 295 bne,pt %icc, 1b 296 EX_ST(STORE(stb, %g2, %o0 - 0x01)) 2972: 298 and %o1, 0x7, %g1 299 brz,pn %g1, .Lmedium_noprefetch 300 sll %g1, 3, %g1 301 mov 64, %g2 302 sub %g2, %g1, %g2 303 andn %o1, 0x7, %o1 304 EX_LD(LOAD(ldx, %o1 + 0x00, %o4)) 305 sllx %o4, %g1, %o4 306 andn %o2, 0x08 - 1, %o5 307 sub %o2, %o5, %o2 3081: EX_LD(LOAD(ldx, %o1 + 0x08, %g3)) 309 add %o1, 0x08, %o1 310 subcc %o5, 0x08, %o5 311 srlx %g3, %g2, GLOBAL_SPARE 312 or GLOBAL_SPARE, %o4, GLOBAL_SPARE 313 EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x00)) 314 add %o0, 0x08, %o0 315 bne,pt %icc, 1b 316 sllx %g3, %g1, %o4 317 srl %g1, 3, %g1 318 add %o1, %g1, %o1 319 brz,pn %o2, .Lexit 320 nop 321 ba,pt %icc, .Lsmall_unaligned 322 323.Ltiny: 324 EX_LD(LOAD(ldub, %o1 + 0x00, %g1)) 325 subcc %o2, 1, %o2 326 be,pn %icc, .Lexit 327 EX_ST(STORE(stb, %g1, %o0 + 0x00)) 328 EX_LD(LOAD(ldub, %o1 + 0x01, %g1)) 329 subcc %o2, 1, %o2 330 be,pn %icc, .Lexit 331 EX_ST(STORE(stb, %g1, %o0 + 0x01)) 332 EX_LD(LOAD(ldub, %o1 + 0x02, %g1)) 333 ba,pt %icc, .Lexit 334 EX_ST(STORE(stb, %g1, %o0 + 0x02)) 335 336.Lsmall: 337 andcc %g2, 0x3, %g0 338 bne,pn %icc, .Lsmall_unaligned 339 andn %o2, 0x4 - 1, %o5 340 sub %o2, %o5, %o2 3411: 342 EX_LD(LOAD(lduw, %o1 + 0x00, %g1)) 343 add %o1, 0x04, %o1 344 subcc %o5, 0x04, %o5 345 add %o0, 0x04, %o0 346 bne,pt %icc, 1b 347 EX_ST(STORE(stw, %g1, %o0 - 0x04)) 348 brz,pt %o2, .Lexit 349 nop 350 ba,a,pt %icc, .Ltiny 351 352.Lsmall_unaligned: 3531: EX_LD(LOAD(ldub, %o1 + 0x00, %g1)) 354 add %o1, 1, %o1 355 add %o0, 1, %o0 356 subcc %o2, 1, %o2 357 bne,pt %icc, 1b 358 EX_ST(STORE(stb, %g1, %o0 - 0x01)) 359 ba,a,pt %icc, .Lexit 360 .size FUNC_NAME, .-FUNC_NAME 361