Skip to content

V
V8
Switch branch/tag
Find file
Normal viewHistoryPermalink
codegen-mips.cc 23.4 KB
Newer Older
yangguo@chromium.org's avatar
MIPS: Some assembler-level optimizations on ARM.
6ab232b1
 
yangguo@chromium.org committed
1 
// Copyright 2012 the V8 project authors. All rights reserved.
bmeurer@chromium.org's avatar
Bulk update of Google copyright headers in source files.
d4b533d4
 
bmeurer@chromium.org committed
2 3 
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
sgjesse@chromium.org's avatar
MIPS port initial commit
a6a7c75a
 
sgjesse@chromium.org committed
4
jkummerow@chromium.org's avatar
Clean up the usage of V8_TARGET_ARCH_${arch} and V8_HOST_ARCH_${arch}
c7a9bffb
 
jkummerow@chromium.org committed
5 
#if V8_TARGET_ARCH_MIPS
mark@chromium.org's avatar
Allow build-time selection between ia32 and x86_64 in the GYP/Xcode Mac
e9b5d785
 
mark@chromium.org committed
6
jochen's avatar
Replace SmartPointer<T> with unique_ptr<T>
0a6ccaf2
 
jochen committed
7 8 
#include <memory>
jochen@chromium.org's avatar
Use full include paths everywhere
56a486c3
 
jochen@chromium.org committed
9 10 11 
#include "src/codegen.h"
#include "src/macro-assembler.h"
#include "src/mips/simulator-mips.h"
sgjesse@chromium.org's avatar
MIPS port initial commit
a6a7c75a
 
sgjesse@chromium.org committed
12 13 14 15 

namespace v8 {
namespace internal {
jkummerow@chromium.org's avatar
MIPS: Faster implementation of Math.exp()
b43ec511
 
jkummerow@chromium.org committed
16 17 
#define __ masm.
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
18 
#if defined(V8_HOST_ARCH_MIPS)
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
19
jochen's avatar
Initialize fast memmove methods in the Isolate's ctor
b93e4d2c
 
jochen committed
20 21 
MemCopyUint8Function CreateMemCopyUint8Function(Isolate* isolate,
                                                MemCopyUint8Function stub) {
dusan.milosavljevic@imgtec.com's avatar
Reland "MIPS: Add support for arch. revision 6 to mips32 port."
8804736b
 
dusan.milosavljevic@imgtec.com committed
22 23 
#if defined(USE_SIMULATOR) || defined(_MIPS_ARCH_MIPS32R6) || \
    defined(_MIPS_ARCH_MIPS32RX)
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
24 25 
  return stub;
#else
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
26 
  size_t allocated = 0;
Bill Budge's avatar
[Memory] Create memory management API in v8::internal.
a449f09f
 
Bill Budge committed
27 
  byte* buffer = AllocatePage(isolate->heap()->GetRandomMmapAddr(), &allocated);
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
28 
  if (buffer == nullptr) return nullptr;
Bill Budge's avatar
Revert "[Memory] Use OS::Allocate for all OS memory allocations."
1ea3fd2e
 
Bill Budge committed
29
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
30 
  MacroAssembler masm(isolate, buffer, static_cast<int>(allocated),
Jakob Gruber's avatar
Revert "Reland "[Memory] Use OS::Allocate for all OS memory allocations.""
8122afa7
 
Jakob Gruber committed
31 32 
                      CodeObjectRequired::kNo);
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
33 34 
  // This code assumes that cache lines are 32 bytes and if the cache line is
  // larger it will not work correctly.
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
35 36 37 38 39 40 41 42 
  {
    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
improve grammar in comment
6e5077a4
 
thakis committed
43 44 
    // 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
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
45 46 47 
    // the kPrefHintPrepareForStore hint is used, the code will not work
    // correctly.
    uint32_t max_pref_size = 128;
bmeurer@chromium.org's avatar
Rename ASSERT* to DCHECK*.
d07a2eb8
 
bmeurer@chromium.org committed
48 
    DCHECK(pref_chunk < max_pref_size);
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
49 50 51 52 53 54 55 56 57 58 59 60 

    // 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.
bmeurer@chromium.org's avatar
Rename ASSERT* to DCHECK*.
d07a2eb8
 
bmeurer@chromium.org committed
61 
    DCHECK(pref_hint_store != kPrefHintPrepareForStore ||
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 
           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(t2, a2, 2 * loadstore_chunk);
    __ bne(t2, 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.
plind44@gmail.com's avatar
MIPS: Add big-endian support for MIPS.
5a016958
 
plind44@gmail.com committed
83 84 85 86 87 88 89 90 91 92 93 
    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);
    }
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
94 95 96 97 98 
    // 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);
Mathias Bynens's avatar
Normalize casing of hexadecimal digits
822be9b2
 
Mathias Bynens committed
99 
    __ andi(t8, a2, 0x3F);
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 
    __ 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 "t0-32" address. This means:
    // for x=128 the last "safe" a0 address is "t0-160". Alternatively, for
    // x=64 the last "safe" a0 address is "t0-96". In the current version we
    // will use "pref hint, 128(a0)", so "t0-160" is the limit.
    if (pref_hint_store == kPrefHintPrepareForStore) {
      __ addu(t0, a0, a2);  // t0 is the "past the end" address.
      __ Subu(t9, t0, 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);
    __ lw(t0, MemOperand(a1));

    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));
    }
    __ lw(t1, MemOperand(a1, 1, 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(&skip_pref);
    __ lw(t2, MemOperand(a1, 2, loadstore_chunk));
    __ lw(t3, MemOperand(a1, 3, loadstore_chunk));
    __ lw(t4, MemOperand(a1, 4, loadstore_chunk));
    __ lw(t5, MemOperand(a1, 5, loadstore_chunk));
    __ lw(t6, MemOperand(a1, 6, loadstore_chunk));
    __ lw(t7, MemOperand(a1, 7, loadstore_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk));

    __ sw(t0, MemOperand(a0));
    __ sw(t1, MemOperand(a0, 1, loadstore_chunk));
    __ sw(t2, MemOperand(a0, 2, loadstore_chunk));
    __ sw(t3, MemOperand(a0, 3, loadstore_chunk));
    __ sw(t4, MemOperand(a0, 4, loadstore_chunk));
    __ sw(t5, MemOperand(a0, 5, loadstore_chunk));
    __ sw(t6, MemOperand(a0, 6, loadstore_chunk));
    __ sw(t7, MemOperand(a0, 7, loadstore_chunk));

    __ lw(t0, MemOperand(a1, 8, loadstore_chunk));
    __ lw(t1, MemOperand(a1, 9, loadstore_chunk));
    __ lw(t2, MemOperand(a1, 10, loadstore_chunk));
    __ lw(t3, MemOperand(a1, 11, loadstore_chunk));
    __ lw(t4, MemOperand(a1, 12, loadstore_chunk));
    __ lw(t5, MemOperand(a1, 13, loadstore_chunk));
    __ lw(t6, MemOperand(a1, 14, loadstore_chunk));
    __ lw(t7, MemOperand(a1, 15, loadstore_chunk));
    __ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk));

    __ sw(t0, MemOperand(a0, 8, loadstore_chunk));
    __ sw(t1, MemOperand(a0, 9, loadstore_chunk));
    __ sw(t2, MemOperand(a0, 10, loadstore_chunk));
    __ sw(t3, MemOperand(a0, 11, loadstore_chunk));
    __ sw(t4, MemOperand(a0, 12, loadstore_chunk));
    __ sw(t5, MemOperand(a0, 13, loadstore_chunk));
    __ sw(t6, MemOperand(a0, 14, loadstore_chunk));
    __ sw(t7, MemOperand(a0, 15, loadstore_chunk));
    __ 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));
Mathias Bynens's avatar
Normalize casing of hexadecimal digits
822be9b2
 
Mathias Bynens committed
182 
    __ andi(t8, a2, 0x1F);
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 
    __ beq(a2, t8, &chk1w);  // Less than 32?
    __ nop();  // In delay slot.
    __ lw(t0, MemOperand(a1));
    __ lw(t1, MemOperand(a1, 1, loadstore_chunk));
    __ lw(t2, MemOperand(a1, 2, loadstore_chunk));
    __ lw(t3, MemOperand(a1, 3, loadstore_chunk));
    __ lw(t4, MemOperand(a1, 4, loadstore_chunk));
    __ lw(t5, MemOperand(a1, 5, loadstore_chunk));
    __ lw(t6, MemOperand(a1, 6, loadstore_chunk));
    __ lw(t7, MemOperand(a1, 7, loadstore_chunk));
    __ addiu(a1, a1, 8 * loadstore_chunk);
    __ sw(t0, MemOperand(a0));
    __ sw(t1, MemOperand(a0, 1, loadstore_chunk));
    __ sw(t2, MemOperand(a0, 2, loadstore_chunk));
    __ sw(t3, MemOperand(a0, 3, loadstore_chunk));
    __ sw(t4, MemOperand(a0, 4, loadstore_chunk));
    __ sw(t5, MemOperand(a0, 5, loadstore_chunk));
    __ sw(t6, MemOperand(a0, 6, loadstore_chunk));
    __ sw(t7, MemOperand(a0, 7, loadstore_chunk));
    __ 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
Julien Brianceau's avatar
Fix common misspellings
b41f857b
 
Julien Brianceau committed
208 
    // and a1 until a0 equals a3.
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 
    __ 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);
    __ lw(t3, MemOperand(a1));
    __ addiu(a0, a0, loadstore_chunk);
    __ addiu(a1, a1, loadstore_chunk);
    __ bne(a0, a3, &wordCopy_loop);
    __ sw(t3, MemOperand(a0, -1, loadstore_chunk));  // In delay slot.

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

    __ bind(&lastbloop);
    __ lb(v1, MemOperand(a1));
    __ addiu(a0, a0, 1);
    __ addiu(a1, a1, 1);
    __ bne(a0, a3, &lastbloop);
    __ sb(v1, MemOperand(a0, -1));  // In delay slot.

    __ 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.
plind44@gmail.com's avatar
MIPS: Add big-endian support for MIPS.
5a016958
 
plind44@gmail.com committed
245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 
    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);
    }
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
260 261 262 263 264 265 

    // 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);
Mathias Bynens's avatar
Normalize casing of hexadecimal digits
822be9b2
 
Mathias Bynens committed
266 
    __ andi(t8, a2, 0x3F);
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 
    __ beq(a2, t8, &ua_chkw);
    __ subu(a3, a2, t8);  // In delay slot.
    __ addu(a3, a0, a3);

    if (pref_hint_store == kPrefHintPrepareForStore) {
      __ addu(t0, a0, a2);
      __ Subu(t9, t0, 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);
    __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));
plind44@gmail.com's avatar
MIPS: Add big-endian support for MIPS.
5a016958
 
plind44@gmail.com committed
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 
    if (kArchEndian == kLittle) {
      __ lwr(t0, MemOperand(a1));
      __ lwr(t1, MemOperand(a1, 1, loadstore_chunk));
      __ lwr(t2, 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(t3, 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(t4, MemOperand(a1, 4, loadstore_chunk));
      __ lwr(t5, MemOperand(a1, 5, loadstore_chunk));
      __ lwr(t6, MemOperand(a1, 6, loadstore_chunk));
      __ lwr(t7, MemOperand(a1, 7, loadstore_chunk));
      __ lwl(t0,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t1,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t2,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t3,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t4,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t5,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t6,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t7,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ lwl(t0, MemOperand(a1));
      __ lwl(t1, MemOperand(a1, 1, loadstore_chunk));
      __ lwl(t2, 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(t3, 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(t4, MemOperand(a1, 4, loadstore_chunk));
      __ lwl(t5, MemOperand(a1, 5, loadstore_chunk));
      __ lwl(t6, MemOperand(a1, 6, loadstore_chunk));
      __ lwl(t7, MemOperand(a1, 7, loadstore_chunk));
      __ lwr(t0,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t1,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t2,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t3,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t4,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t5,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t6,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t7,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
358 359 360 361 362 363 364 365 366 367 
    }
    __ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk));
    __ sw(t0, MemOperand(a0));
    __ sw(t1, MemOperand(a0, 1, loadstore_chunk));
    __ sw(t2, MemOperand(a0, 2, loadstore_chunk));
    __ sw(t3, MemOperand(a0, 3, loadstore_chunk));
    __ sw(t4, MemOperand(a0, 4, loadstore_chunk));
    __ sw(t5, MemOperand(a0, 5, loadstore_chunk));
    __ sw(t6, MemOperand(a0, 6, loadstore_chunk));
    __ sw(t7, MemOperand(a0, 7, loadstore_chunk));
plind44@gmail.com's avatar
MIPS: Add big-endian support for MIPS.
5a016958
 
plind44@gmail.com committed
368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 
    if (kArchEndian == kLittle) {
      __ lwr(t0, MemOperand(a1, 8, loadstore_chunk));
      __ lwr(t1, MemOperand(a1, 9, loadstore_chunk));
      __ lwr(t2, MemOperand(a1, 10, loadstore_chunk));
      __ lwr(t3, MemOperand(a1, 11, loadstore_chunk));
      __ lwr(t4, MemOperand(a1, 12, loadstore_chunk));
      __ lwr(t5, MemOperand(a1, 13, loadstore_chunk));
      __ lwr(t6, MemOperand(a1, 14, loadstore_chunk));
      __ lwr(t7, MemOperand(a1, 15, loadstore_chunk));
      __ lwl(t0,
             MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t1,
             MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t2,
             MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t3,
             MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t4,
             MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t5,
             MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t6,
             MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t7,
             MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ lwl(t0, MemOperand(a1, 8, loadstore_chunk));
      __ lwl(t1, MemOperand(a1, 9, loadstore_chunk));
      __ lwl(t2, MemOperand(a1, 10, loadstore_chunk));
      __ lwl(t3, MemOperand(a1, 11, loadstore_chunk));
      __ lwl(t4, MemOperand(a1, 12, loadstore_chunk));
      __ lwl(t5, MemOperand(a1, 13, loadstore_chunk));
      __ lwl(t6, MemOperand(a1, 14, loadstore_chunk));
      __ lwl(t7, MemOperand(a1, 15, loadstore_chunk));
      __ lwr(t0,
             MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t1,
             MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t2,
             MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t3,
             MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t4,
             MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t5,
             MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t6,
             MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t7,
             MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one));
    }
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 
    __ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk));
    __ sw(t0, MemOperand(a0, 8, loadstore_chunk));
    __ sw(t1, MemOperand(a0, 9, loadstore_chunk));
    __ sw(t2, MemOperand(a0, 10, loadstore_chunk));
    __ sw(t3, MemOperand(a0, 11, loadstore_chunk));
    __ sw(t4, MemOperand(a0, 12, loadstore_chunk));
    __ sw(t5, MemOperand(a0, 13, loadstore_chunk));
    __ sw(t6, MemOperand(a0, 14, loadstore_chunk));
    __ sw(t7, MemOperand(a0, 15, loadstore_chunk));
    __ 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));
Mathias Bynens's avatar
Normalize casing of hexadecimal digits
822be9b2
 
Mathias Bynens committed
438 
    __ andi(t8, a2, 0x1F);
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
439 440 441 

    __ beq(a2, t8, &ua_chk1w);
    __ nop();  // In delay slot.
plind44@gmail.com's avatar
MIPS: Add big-endian support for MIPS.
5a016958
 
plind44@gmail.com committed
442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 
    if (kArchEndian == kLittle) {
      __ lwr(t0, MemOperand(a1));
      __ lwr(t1, MemOperand(a1, 1, loadstore_chunk));
      __ lwr(t2, MemOperand(a1, 2, loadstore_chunk));
      __ lwr(t3, MemOperand(a1, 3, loadstore_chunk));
      __ lwr(t4, MemOperand(a1, 4, loadstore_chunk));
      __ lwr(t5, MemOperand(a1, 5, loadstore_chunk));
      __ lwr(t6, MemOperand(a1, 6, loadstore_chunk));
      __ lwr(t7, MemOperand(a1, 7, loadstore_chunk));
      __ lwl(t0,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t1,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t2,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t3,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t4,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t5,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t6,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwl(t7,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
    } else {
      __ lwl(t0, MemOperand(a1));
      __ lwl(t1, MemOperand(a1, 1, loadstore_chunk));
      __ lwl(t2, MemOperand(a1, 2, loadstore_chunk));
      __ lwl(t3, MemOperand(a1, 3, loadstore_chunk));
      __ lwl(t4, MemOperand(a1, 4, loadstore_chunk));
      __ lwl(t5, MemOperand(a1, 5, loadstore_chunk));
      __ lwl(t6, MemOperand(a1, 6, loadstore_chunk));
      __ lwl(t7, MemOperand(a1, 7, loadstore_chunk));
      __ lwr(t0,
             MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t1,
             MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t2,
             MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t3,
             MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t4,
             MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t5,
             MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t6,
             MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
      __ lwr(t7,
             MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
    }
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 
    __ addiu(a1, a1, 8 * loadstore_chunk);
    __ sw(t0, MemOperand(a0));
    __ sw(t1, MemOperand(a0, 1, loadstore_chunk));
    __ sw(t2, MemOperand(a0, 2, loadstore_chunk));
    __ sw(t3, MemOperand(a0, 3, loadstore_chunk));
    __ sw(t4, MemOperand(a0, 4, loadstore_chunk));
    __ sw(t5, MemOperand(a0, 5, loadstore_chunk));
    __ sw(t6, MemOperand(a0, 6, loadstore_chunk));
    __ sw(t7, MemOperand(a0, 7, loadstore_chunk));
    __ 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);
plind44@gmail.com's avatar
MIPS: Add big-endian support for MIPS.
5a016958
 
plind44@gmail.com committed
513 514 515 516 517 518 519 520 521 
    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));
    }
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 
    __ addiu(a0, a0, loadstore_chunk);
    __ addiu(a1, a1, loadstore_chunk);
    __ bne(a0, a3, &ua_wordCopy_loop);
    __ sw(v1, MemOperand(a0, -1, loadstore_chunk));  // In delay slot.

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

    __ bind(&ua_smallCopy_loop);
    __ lb(v1, MemOperand(a1));
    __ addiu(a0, a0, 1);
    __ addiu(a1, a1, 1);
    __ bne(a0, a3, &ua_smallCopy_loop);
    __ sb(v1, MemOperand(a0, -1));  // In delay slot.

    __ jr(ra);
    __ nop();
  }
  CodeDesc desc;
neis's avatar
[compiler] Delay allocation of code-embedded heap numbers.
659e8f7b
 
neis committed
543 
  masm.GetCode(isolate, &desc);
neis's avatar
Disentangle assembler from isolate.
94b088ca
 
neis committed
544 
  DCHECK(!RelocInfo::RequiresRelocation(isolate, desc));
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
545
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
546 
  Assembler::FlushICache(isolate, buffer, allocated);
Bill Budge's avatar
[Memory] Add PageAllocator concept to v8::Platform.
699144a2
 
Bill Budge committed
547 
  CHECK(SetPermissions(buffer, allocated, PageAllocator::kReadExecute));
jochen@chromium.org's avatar
Reland 21502 - "Move OS::MemCopy and OS::MemMove out of platform to utils"
84e078e5
 
jochen@chromium.org committed
548 
  return FUNCTION_CAST<MemCopyUint8Function>(buffer);
plind44@gmail.com's avatar
MIPS: Faster memcpy.
17766404
 
plind44@gmail.com committed
549 550 551 552 
#endif
}
#endif
jochen's avatar
Lazily initialize fast_sqrt() and pass an Isolate parameter to it
27001ca6
 
jochen committed
553 
UnaryMathFunctionWithIsolate CreateSqrtFunction(Isolate* isolate) {
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
554 
#if defined(USE_SIMULATOR)
jochen's avatar
Lazily initialize fast_sqrt() and pass an Isolate parameter to it
27001ca6
 
jochen committed
555 
  return nullptr;
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
556 
#else
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
557 
  size_t allocated = 0;
Bill Budge's avatar
[Memory] Create memory management API in v8::internal.
a449f09f
 
Bill Budge committed
558 
  byte* buffer = AllocatePage(isolate->heap()->GetRandomMmapAddr(), &allocated);
jochen's avatar
Lazily initialize fast_sqrt() and pass an Isolate parameter to it
27001ca6
 
jochen committed
559 
  if (buffer == nullptr) return nullptr;
jkummerow@chromium.org's avatar
MIPS: Faster implementation of Math.exp()
b43ec511
 
jkummerow@chromium.org committed
560
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
561 
  MacroAssembler masm(isolate, buffer, static_cast<int>(allocated),
jochen's avatar
Make whether or not a Code object should be created by masm explicit
aa9cfc82
 
jochen committed
562 
                      CodeObjectRequired::kNo);
jkummerow@chromium.org's avatar
MIPS: Faster implementation of Math.exp()
b43ec511
 
jkummerow@chromium.org committed
563
svenpanne@chromium.org's avatar
Demystified floating point ABI function names.
a5535f58
 
svenpanne@chromium.org committed
564 
  __ MovFromFloatParameter(f12);
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
565 
  __ sqrt_d(f0, f12);
svenpanne@chromium.org's avatar
Demystified floating point ABI function names.
a5535f58
 
svenpanne@chromium.org committed
566 
  __ MovToFloatResult(f0);
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
567 568 569 
  __ Ret();

  CodeDesc desc;
neis's avatar
[compiler] Delay allocation of code-embedded heap numbers.
659e8f7b
 
neis committed
570 
  masm.GetCode(isolate, &desc);
neis's avatar
Disentangle assembler from isolate.
94b088ca
 
neis committed
571 
  DCHECK(!RelocInfo::RequiresRelocation(isolate, desc));
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
572
Bill Budge's avatar
Reland "[Memory] Use OS::Allocate for all OS memory allocations."
adc52af5
 
Bill Budge committed
573 
  Assembler::FlushICache(isolate, buffer, allocated);
Bill Budge's avatar
[Memory] Add PageAllocator concept to v8::Platform.
699144a2
 
Bill Budge committed
574 
  CHECK(SetPermissions(buffer, allocated, PageAllocator::kReadExecute));
jochen's avatar
Lazily initialize fast_sqrt() and pass an Isolate parameter to it
27001ca6
 
jochen committed
575 
  return FUNCTION_CAST<UnaryMathFunctionWithIsolate>(buffer);
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
576 
#endif
yangguo@chromium.org's avatar
Ensure consistency of Math.sqrt on Intel platforms.
7659beaf
 
yangguo@chromium.org committed
577 578 
}
plind44@gmail.com's avatar
MIPS: Implement sqrt in inline assembly.
e3f474e1
 
plind44@gmail.com committed
579 580 
#undef __
erikcorry's avatar
Cosmetic changes to tests to make it easier to concatenate them.
4f5337a2
 
erikcorry committed
581 582 
}  // namespace internal
}  // namespace v8
mark@chromium.org's avatar
Allow build-time selection between ia32 and x86_64 in the GYP/Xcode Mac
e9b5d785
 
mark@chromium.org committed
583 584 

#endif  // V8_TARGET_ARCH_MIPS