codegen-mips64.cc 23.4 KB
Newer Older
1 2 3 4 5 6
// Copyright 2012 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#if V8_TARGET_ARCH_MIPS64

7 8
#include <memory>

9 10 11 12 13 14 15 16 17 18 19 20
#include "src/codegen.h"
#include "src/macro-assembler.h"
#include "src/mips64/simulator-mips64.h"

namespace v8 {
namespace internal {


#define __ masm.


#if defined(V8_HOST_ARCH_MIPS)
21 22
MemCopyUint8Function CreateMemCopyUint8Function(Isolate* isolate,
                                                MemCopyUint8Function stub) {
23 24 25 26 27 28 29
#if defined(USE_SIMULATOR)
  return stub;
#else

  size_t actual_size;
  byte* buffer =
      static_cast<byte*>(base::OS::Allocate(3 * KB, &actual_size, true));
30
  if (buffer == nullptr) return stub;
31 32 33

  // This code assumes that cache lines are 32 bytes and if the cache line is
  // larger it will not work correctly.
34
  MacroAssembler masm(isolate, buffer, static_cast<int>(actual_size),
35
                      CodeObjectRequired::kNo);
36 37 38 39 40 41 42 43 44

  {
    Label lastb, unaligned, aligned, chkw,
          loop16w, chk1w, wordCopy_loop, skip_pref, lastbloop,
          leave, ua_chk16w, ua_loop16w, ua_skip_pref, ua_chkw,
          ua_chk1w, ua_wordCopy_loop, ua_smallCopy, ua_smallCopy_loop;

    // The size of each prefetch.
    uint32_t pref_chunk = 32;
thakis's avatar
thakis committed
45 46
    // The maximum size of a prefetch, it must not be less than pref_chunk.
    // If the real size of a prefetch is greater than max_pref_size and
47 48 49
    // the kPrefHintPrepareForStore hint is used, the code will not work
    // correctly.
    uint32_t max_pref_size = 128;
50
    DCHECK(pref_chunk < max_pref_size);
51 52 53 54 55 56 57 58 59 60 61 62

    // pref_limit is set based on the fact that we never use an offset
    // greater then 5 on a store pref and that a single pref can
    // never be larger then max_pref_size.
    uint32_t pref_limit = (5 * pref_chunk) + max_pref_size;
    int32_t pref_hint_load = kPrefHintLoadStreamed;
    int32_t pref_hint_store = kPrefHintPrepareForStore;
    uint32_t loadstore_chunk = 4;

    // The initial prefetches may fetch bytes that are before the buffer being
    // copied. Start copies with an offset of 4 so avoid this situation when
    // using kPrefHintPrepareForStore.
63
    DCHECK(pref_hint_store != kPrefHintPrepareForStore ||
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
           pref_chunk * 4 >= max_pref_size);
    // If the size is less than 8, go to lastb. Regardless of size,
    // copy dst pointer to v0 for the retuen value.
    __ slti(a6, a2, 2 * loadstore_chunk);
    __ bne(a6, zero_reg, &lastb);
    __ mov(v0, a0);  // In delay slot.

    // If src and dst have different alignments, go to unaligned, if they
    // have the same alignment (but are not actually aligned) do a partial
    // load/store to make them aligned. If they are both already aligned
    // we can start copying at aligned.
    __ xor_(t8, a1, a0);
    __ andi(t8, t8, loadstore_chunk - 1);  // t8 is a0/a1 word-displacement.
    __ bne(t8, zero_reg, &unaligned);
    __ subu(a3, zero_reg, a0);  // In delay slot.

    __ andi(a3, a3, loadstore_chunk - 1);  // Copy a3 bytes to align a0/a1.
    __ beq(a3, zero_reg, &aligned);  // Already aligned.
    __ subu(a2, a2, a3);  // In delay slot. a2 is the remining bytes count.

84 85 86 87 88 89 90 91 92 93 94
    if (kArchEndian == kLittle) {
      __ lwr(t8, MemOperand(a1));
      __ addu(a1, a1, a3);
      __ swr(t8, MemOperand(a0));
      __ addu(a0, a0, a3);
    } else {
      __ lwl(t8, MemOperand(a1));
      __ addu(a1, a1, a3);
      __ swl(t8, MemOperand(a0));
      __ addu(a0, a0, a3);
    }
95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126

    // Now dst/src are both aligned to (word) aligned addresses. Set a2 to
    // count how many bytes we have to copy after all the 64 byte chunks are
    // copied and a3 to the dst pointer after all the 64 byte chunks have been
    // copied. We will loop, incrementing a0 and a1 until a0 equals a3.
    __ bind(&aligned);
    __ andi(t8, a2, 0x3f);
    __ beq(a2, t8, &chkw);  // Less than 64?
    __ subu(a3, a2, t8);  // In delay slot.
    __ addu(a3, a0, a3);  // Now a3 is the final dst after loop.

    // When in the loop we prefetch with kPrefHintPrepareForStore hint,
    // in this case the a0+x should be past the "a4-32" address. This means:
    // for x=128 the last "safe" a0 address is "a4-160". Alternatively, for
    // x=64 the last "safe" a0 address is "a4-96". In the current version we
    // will use "pref hint, 128(a0)", so "a4-160" is the limit.
    if (pref_hint_store == kPrefHintPrepareForStore) {
      __ addu(a4, a0, a2);  // a4 is the "past the end" address.
      __ Subu(t9, a4, pref_limit);  // t9 is the "last safe pref" address.
    }

    __ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 1 * pref_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 2 * pref_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));

    if (pref_hint_store != kPrefHintPrepareForStore) {
      __ Pref(pref_hint_store, MemOperand(a0, 1 * pref_chunk));
      __ Pref(pref_hint_store, MemOperand(a0, 2 * pref_chunk));
      __ Pref(pref_hint_store, MemOperand(a0, 3 * pref_chunk));
    }
    __ bind(&loop16w);
127
    __ Lw(a4, MemOperand(a1));
128 129 130 131 132

    if (pref_hint_store == kPrefHintPrepareForStore) {
      __ sltu(v1, t9, a0);  // If a0 > t9, don't use next prefetch.
      __ Branch(USE_DELAY_SLOT, &skip_pref, gt, v1, Operand(zero_reg));
    }
133
    __ Lw(a5, MemOperand(a1, 1, loadstore_chunk));  // Maybe in delay slot.
134 135 136 137 138

    __ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk));
    __ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk));

    __ bind(&skip_pref);
139 140 141 142 143 144
    __ Lw(a6, MemOperand(a1, 2, loadstore_chunk));
    __ Lw(a7, MemOperand(a1, 3, loadstore_chunk));
    __ Lw(t0, MemOperand(a1, 4, loadstore_chunk));
    __ Lw(t1, MemOperand(a1, 5, loadstore_chunk));
    __ Lw(t2, MemOperand(a1, 6, loadstore_chunk));
    __ Lw(t3, MemOperand(a1, 7, loadstore_chunk));
145 146
    __ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk));

147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
    __ Sw(a4, MemOperand(a0));
    __ Sw(a5, MemOperand(a0, 1, loadstore_chunk));
    __ Sw(a6, MemOperand(a0, 2, loadstore_chunk));
    __ Sw(a7, MemOperand(a0, 3, loadstore_chunk));
    __ Sw(t0, MemOperand(a0, 4, loadstore_chunk));
    __ Sw(t1, MemOperand(a0, 5, loadstore_chunk));
    __ Sw(t2, MemOperand(a0, 6, loadstore_chunk));
    __ Sw(t3, MemOperand(a0, 7, loadstore_chunk));

    __ Lw(a4, MemOperand(a1, 8, loadstore_chunk));
    __ Lw(a5, MemOperand(a1, 9, loadstore_chunk));
    __ Lw(a6, MemOperand(a1, 10, loadstore_chunk));
    __ Lw(a7, MemOperand(a1, 11, loadstore_chunk));
    __ Lw(t0, MemOperand(a1, 12, loadstore_chunk));
    __ Lw(t1, MemOperand(a1, 13, loadstore_chunk));
    __ Lw(t2, MemOperand(a1, 14, loadstore_chunk));
    __ Lw(t3, MemOperand(a1, 15, loadstore_chunk));
164 165
    __ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk));

166 167 168 169 170 171 172 173
    __ Sw(a4, MemOperand(a0, 8, loadstore_chunk));
    __ Sw(a5, MemOperand(a0, 9, loadstore_chunk));
    __ Sw(a6, MemOperand(a0, 10, loadstore_chunk));
    __ Sw(a7, MemOperand(a0, 11, loadstore_chunk));
    __ Sw(t0, MemOperand(a0, 12, loadstore_chunk));
    __ Sw(t1, MemOperand(a0, 13, loadstore_chunk));
    __ Sw(t2, MemOperand(a0, 14, loadstore_chunk));
    __ Sw(t3, MemOperand(a0, 15, loadstore_chunk));
174 175 176 177 178 179 180 181 182 183 184 185 186
    __ addiu(a0, a0, 16 * loadstore_chunk);
    __ bne(a0, a3, &loop16w);
    __ addiu(a1, a1, 16 * loadstore_chunk);  // In delay slot.
    __ mov(a2, t8);

    // Here we have src and dest word-aligned but less than 64-bytes to go.
    // Check for a 32 bytes chunk and copy if there is one. Otherwise jump
    // down to chk1w to handle the tail end of the copy.
    __ bind(&chkw);
    __ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk));
    __ andi(t8, a2, 0x1f);
    __ beq(a2, t8, &chk1w);  // Less than 32?
    __ nop();  // In delay slot.
187 188 189 190 191 192 193 194
    __ Lw(a4, MemOperand(a1));
    __ Lw(a5, MemOperand(a1, 1, loadstore_chunk));
    __ Lw(a6, MemOperand(a1, 2, loadstore_chunk));
    __ Lw(a7, MemOperand(a1, 3, loadstore_chunk));
    __ Lw(t0, MemOperand(a1, 4, loadstore_chunk));
    __ Lw(t1, MemOperand(a1, 5, loadstore_chunk));
    __ Lw(t2, MemOperand(a1, 6, loadstore_chunk));
    __ Lw(t3, MemOperand(a1, 7, loadstore_chunk));
195
    __ addiu(a1, a1, 8 * loadstore_chunk);
196 197 198 199 200 201 202 203
    __ Sw(a4, MemOperand(a0));
    __ Sw(a5, MemOperand(a0, 1, loadstore_chunk));
    __ Sw(a6, MemOperand(a0, 2, loadstore_chunk));
    __ Sw(a7, MemOperand(a0, 3, loadstore_chunk));
    __ Sw(t0, MemOperand(a0, 4, loadstore_chunk));
    __ Sw(t1, MemOperand(a0, 5, loadstore_chunk));
    __ Sw(t2, MemOperand(a0, 6, loadstore_chunk));
    __ Sw(t3, MemOperand(a0, 7, loadstore_chunk));
204 205 206 207 208 209
    __ addiu(a0, a0, 8 * loadstore_chunk);

    // Here we have less than 32 bytes to copy. Set up for a loop to copy
    // one word at a time. Set a2 to count how many bytes we have to copy
    // after all the word chunks are copied and a3 to the dst pointer after
    // all the word chunks have been copied. We will loop, incrementing a0
210
    // and a1 until a0 equals a3.
211 212 213 214 215 216 217
    __ bind(&chk1w);
    __ andi(a2, t8, loadstore_chunk - 1);
    __ beq(a2, t8, &lastb);
    __ subu(a3, t8, a2);  // In delay slot.
    __ addu(a3, a0, a3);

    __ bind(&wordCopy_loop);
218
    __ Lw(a7, MemOperand(a1));
219 220 221
    __ addiu(a0, a0, loadstore_chunk);
    __ addiu(a1, a1, loadstore_chunk);
    __ bne(a0, a3, &wordCopy_loop);
222
    __ Sw(a7, MemOperand(a0, -1, loadstore_chunk));  // In delay slot.
223 224 225 226 227 228

    __ bind(&lastb);
    __ Branch(&leave, le, a2, Operand(zero_reg));
    __ addu(a3, a0, a2);

    __ bind(&lastbloop);
229
    __ Lb(v1, MemOperand(a1));
230 231 232
    __ addiu(a0, a0, 1);
    __ addiu(a1, a1, 1);
    __ bne(a0, a3, &lastbloop);
233
    __ Sb(v1, MemOperand(a0, -1));  // In delay slot.
234 235 236 237 238 239 240 241 242 243 244 245 246

    __ bind(&leave);
    __ jr(ra);
    __ nop();

    // Unaligned case. Only the dst gets aligned so we need to do partial
    // loads of the source followed by normal stores to the dst (once we
    // have aligned the destination).
    __ bind(&unaligned);
    __ andi(a3, a3, loadstore_chunk - 1);  // Copy a3 bytes to align a0/a1.
    __ beq(a3, zero_reg, &ua_chk16w);
    __ subu(a2, a2, a3);  // In delay slot.

247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
    if (kArchEndian == kLittle) {
      __ lwr(v1, MemOperand(a1));
      __ lwl(v1,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ addu(a1, a1, a3);
      __ swr(v1, MemOperand(a0));
      __ addu(a0, a0, a3);
    } else {
      __ lwl(v1, MemOperand(a1));
      __ lwr(v1,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ addu(a1, a1, a3);
      __ swl(v1, MemOperand(a0));
      __ addu(a0, a0, a3);
    }
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288

    // Now the dst (but not the source) is aligned. Set a2 to count how many
    // bytes we have to copy after all the 64 byte chunks are copied and a3 to
    // the dst pointer after all the 64 byte chunks have been copied. We will
    // loop, incrementing a0 and a1 until a0 equals a3.
    __ bind(&ua_chk16w);
    __ andi(t8, a2, 0x3f);
    __ beq(a2, t8, &ua_chkw);
    __ subu(a3, a2, t8);  // In delay slot.
    __ addu(a3, a0, a3);

    if (pref_hint_store == kPrefHintPrepareForStore) {
      __ addu(a4, a0, a2);
      __ Subu(t9, a4, pref_limit);
    }

    __ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 1 * pref_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 2 * pref_chunk));

    if (pref_hint_store != kPrefHintPrepareForStore) {
      __ Pref(pref_hint_store, MemOperand(a0, 1 * pref_chunk));
      __ Pref(pref_hint_store, MemOperand(a0, 2 * pref_chunk));
      __ Pref(pref_hint_store, MemOperand(a0, 3 * pref_chunk));
    }

    __ bind(&ua_loop16w);
289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
    if (kArchEndian == kLittle) {
      __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));
      __ lwr(a4, MemOperand(a1));
      __ lwr(a5, MemOperand(a1, 1, loadstore_chunk));
      __ lwr(a6, MemOperand(a1, 2, loadstore_chunk));

      if (pref_hint_store == kPrefHintPrepareForStore) {
        __ sltu(v1, t9, a0);
        __ Branch(USE_DELAY_SLOT, &ua_skip_pref, gt, v1, Operand(zero_reg));
      }
      __ lwr(a7, MemOperand(a1, 3, loadstore_chunk));  // Maybe in delay slot.

      __ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk));
      __ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk));

      __ bind(&ua_skip_pref);
      __ lwr(t0, MemOperand(a1, 4, loadstore_chunk));
      __ lwr(t1, MemOperand(a1, 5, loadstore_chunk));
      __ lwr(t2, MemOperand(a1, 6, loadstore_chunk));
      __ lwr(t3, MemOperand(a1, 7, loadstore_chunk));
      __ lwl(a4,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a5,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a6,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a7,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t0,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t1,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t2,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t3,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));
      __ lwl(a4, MemOperand(a1));
      __ lwl(a5, MemOperand(a1, 1, loadstore_chunk));
      __ lwl(a6, MemOperand(a1, 2, loadstore_chunk));

      if (pref_hint_store == kPrefHintPrepareForStore) {
        __ sltu(v1, t9, a0);
        __ Branch(USE_DELAY_SLOT, &ua_skip_pref, gt, v1, Operand(zero_reg));
      }
      __ lwl(a7, MemOperand(a1, 3, loadstore_chunk));  // Maybe in delay slot.

      __ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk));
      __ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk));

      __ bind(&ua_skip_pref);
      __ lwl(t0, MemOperand(a1, 4, loadstore_chunk));
      __ lwl(t1, MemOperand(a1, 5, loadstore_chunk));
      __ lwl(t2, MemOperand(a1, 6, loadstore_chunk));
      __ lwl(t3, MemOperand(a1, 7, loadstore_chunk));
      __ lwr(a4,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a5,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a6,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a7,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t0,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t1,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t2,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t3,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
361 362
    }
    __ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk));
363 364 365 366 367 368 369 370
    __ Sw(a4, MemOperand(a0));
    __ Sw(a5, MemOperand(a0, 1, loadstore_chunk));
    __ Sw(a6, MemOperand(a0, 2, loadstore_chunk));
    __ Sw(a7, MemOperand(a0, 3, loadstore_chunk));
    __ Sw(t0, MemOperand(a0, 4, loadstore_chunk));
    __ Sw(t1, MemOperand(a0, 5, loadstore_chunk));
    __ Sw(t2, MemOperand(a0, 6, loadstore_chunk));
    __ Sw(t3, MemOperand(a0, 7, loadstore_chunk));
371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
    if (kArchEndian == kLittle) {
      __ lwr(a4, MemOperand(a1, 8, loadstore_chunk));
      __ lwr(a5, MemOperand(a1, 9, loadstore_chunk));
      __ lwr(a6, MemOperand(a1, 10, loadstore_chunk));
      __ lwr(a7, MemOperand(a1, 11, loadstore_chunk));
      __ lwr(t0, MemOperand(a1, 12, loadstore_chunk));
      __ lwr(t1, MemOperand(a1, 13, loadstore_chunk));
      __ lwr(t2, MemOperand(a1, 14, loadstore_chunk));
      __ lwr(t3, MemOperand(a1, 15, loadstore_chunk));
      __ lwl(a4,
             MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a5,
             MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a6,
             MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a7,
             MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t0,
             MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t1,
             MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t2,
             MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t3,
             MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ lwl(a4, MemOperand(a1, 8, loadstore_chunk));
      __ lwl(a5, MemOperand(a1, 9, loadstore_chunk));
      __ lwl(a6, MemOperand(a1, 10, loadstore_chunk));
      __ lwl(a7, MemOperand(a1, 11, loadstore_chunk));
      __ lwl(t0, MemOperand(a1, 12, loadstore_chunk));
      __ lwl(t1, MemOperand(a1, 13, loadstore_chunk));
      __ lwl(t2, MemOperand(a1, 14, loadstore_chunk));
      __ lwl(t3, MemOperand(a1, 15, loadstore_chunk));
      __ lwr(a4,
             MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a5,
             MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a6,
             MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a7,
             MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t0,
             MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t1,
             MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t2,
             MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t3,
             MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one));
    }
422
    __ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk));
423 424 425 426 427 428 429 430
    __ Sw(a4, MemOperand(a0, 8, loadstore_chunk));
    __ Sw(a5, MemOperand(a0, 9, loadstore_chunk));
    __ Sw(a6, MemOperand(a0, 10, loadstore_chunk));
    __ Sw(a7, MemOperand(a0, 11, loadstore_chunk));
    __ Sw(t0, MemOperand(a0, 12, loadstore_chunk));
    __ Sw(t1, MemOperand(a0, 13, loadstore_chunk));
    __ Sw(t2, MemOperand(a0, 14, loadstore_chunk));
    __ Sw(t3, MemOperand(a0, 15, loadstore_chunk));
431 432 433 434 435 436 437 438 439 440 441 442 443 444
    __ addiu(a0, a0, 16 * loadstore_chunk);
    __ bne(a0, a3, &ua_loop16w);
    __ addiu(a1, a1, 16 * loadstore_chunk);  // In delay slot.
    __ mov(a2, t8);

    // Here less than 64-bytes. Check for
    // a 32 byte chunk and copy if there is one. Otherwise jump down to
    // ua_chk1w to handle the tail end of the copy.
    __ bind(&ua_chkw);
    __ Pref(pref_hint_load, MemOperand(a1));
    __ andi(t8, a2, 0x1f);

    __ beq(a2, t8, &ua_chk1w);
    __ nop();  // In delay slot.
445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
    if (kArchEndian == kLittle) {
      __ lwr(a4, MemOperand(a1));
      __ lwr(a5, MemOperand(a1, 1, loadstore_chunk));
      __ lwr(a6, MemOperand(a1, 2, loadstore_chunk));
      __ lwr(a7, MemOperand(a1, 3, loadstore_chunk));
      __ lwr(t0, MemOperand(a1, 4, loadstore_chunk));
      __ lwr(t1, MemOperand(a1, 5, loadstore_chunk));
      __ lwr(t2, MemOperand(a1, 6, loadstore_chunk));
      __ lwr(t3, MemOperand(a1, 7, loadstore_chunk));
      __ lwl(a4,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a5,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a6,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(a7,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t0,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t1,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t2,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t3,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ lwl(a4, MemOperand(a1));
      __ lwl(a5, MemOperand(a1, 1, loadstore_chunk));
      __ lwl(a6, MemOperand(a1, 2, loadstore_chunk));
      __ lwl(a7, MemOperand(a1, 3, loadstore_chunk));
      __ lwl(t0, MemOperand(a1, 4, loadstore_chunk));
      __ lwl(t1, MemOperand(a1, 5, loadstore_chunk));
      __ lwl(t2, MemOperand(a1, 6, loadstore_chunk));
      __ lwl(t3, MemOperand(a1, 7, loadstore_chunk));
      __ lwr(a4,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a5,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a6,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(a7,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t0,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t1,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t2,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t3,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
    }
496
    __ addiu(a1, a1, 8 * loadstore_chunk);
497 498 499 500 501 502 503 504
    __ Sw(a4, MemOperand(a0));
    __ Sw(a5, MemOperand(a0, 1, loadstore_chunk));
    __ Sw(a6, MemOperand(a0, 2, loadstore_chunk));
    __ Sw(a7, MemOperand(a0, 3, loadstore_chunk));
    __ Sw(t0, MemOperand(a0, 4, loadstore_chunk));
    __ Sw(t1, MemOperand(a0, 5, loadstore_chunk));
    __ Sw(t2, MemOperand(a0, 6, loadstore_chunk));
    __ Sw(t3, MemOperand(a0, 7, loadstore_chunk));
505 506 507 508 509 510 511 512 513 514 515
    __ addiu(a0, a0, 8 * loadstore_chunk);

    // Less than 32 bytes to copy. Set up for a loop to
    // copy one word at a time.
    __ bind(&ua_chk1w);
    __ andi(a2, t8, loadstore_chunk - 1);
    __ beq(a2, t8, &ua_smallCopy);
    __ subu(a3, t8, a2);  // In delay slot.
    __ addu(a3, a0, a3);

    __ bind(&ua_wordCopy_loop);
516 517 518 519 520 521 522 523 524
    if (kArchEndian == kLittle) {
      __ lwr(v1, MemOperand(a1));
      __ lwl(v1,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ lwl(v1, MemOperand(a1));
      __ lwr(v1,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
    }
525 526 527
    __ addiu(a0, a0, loadstore_chunk);
    __ addiu(a1, a1, loadstore_chunk);
    __ bne(a0, a3, &ua_wordCopy_loop);
528
    __ Sw(v1, MemOperand(a0, -1, loadstore_chunk));  // In delay slot.
529 530 531 532 533 534 535

    // Copy the last 8 bytes.
    __ bind(&ua_smallCopy);
    __ beq(a2, zero_reg, &leave);
    __ addu(a3, a0, a2);  // In delay slot.

    __ bind(&ua_smallCopy_loop);
536
    __ Lb(v1, MemOperand(a1));
537 538 539
    __ addiu(a0, a0, 1);
    __ addiu(a1, a1, 1);
    __ bne(a0, a3, &ua_smallCopy_loop);
540
    __ Sb(v1, MemOperand(a0, -1));  // In delay slot.
541 542 543 544 545

    __ jr(ra);
    __ nop();
  }
  CodeDesc desc;
546
  masm.GetCode(isolte, &desc);
547
  DCHECK(!RelocInfo::RequiresRelocation(isolate, desc));
548

549
  Assembler::FlushICache(isolate, buffer, actual_size);
550
  base::OS::SetReadAndExecutable(buffer, actual_size);
551 552 553 554 555
  return FUNCTION_CAST<MemCopyUint8Function>(buffer);
#endif
}
#endif

556
UnaryMathFunctionWithIsolate CreateSqrtFunction(Isolate* isolate) {
557
#if defined(USE_SIMULATOR)
558
  return nullptr;
559 560 561 562
#else
  size_t actual_size;
  byte* buffer =
      static_cast<byte*>(base::OS::Allocate(1 * KB, &actual_size, true));
563
  if (buffer == nullptr) return nullptr;
564

565
  MacroAssembler masm(isolate, buffer, static_cast<int>(actual_size),
566
                      CodeObjectRequired::kNo);
567 568 569 570 571 572 573

  __ MovFromFloatParameter(f12);
  __ sqrt_d(f0, f12);
  __ MovToFloatResult(f0);
  __ Ret();

  CodeDesc desc;
574
  masm.GetCode(isolate, &desc);
575
  DCHECK(!RelocInfo::RequiresRelocation(isolate, desc));
576

577
  Assembler::FlushICache(isolate, buffer, actual_size);
578
  base::OS::SetReadAndExecutable(buffer, actual_size);
579
  return FUNCTION_CAST<UnaryMathFunctionWithIsolate>(buffer);
580 581 582 583 584
#endif
}

#undef __

585 586
}  // namespace internal
}  // namespace v8
587 588

#endif  // V8_TARGET_ARCH_MIPS64