deoptimizer-ppc.cc 12.2 KB
Newer Older
1 2 3 4 5 6
// Copyright 2014 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.

#include "src/codegen.h"
#include "src/deoptimizer.h"
7
#include "src/full-codegen/full-codegen.h"
8
#include "src/register-configuration.h"
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
#include "src/safepoint-table.h"

namespace v8 {
namespace internal {

const int Deoptimizer::table_entry_size_ = 8;


int Deoptimizer::patch_size() {
#if V8_TARGET_ARCH_PPC64
  const int kCallInstructionSizeInWords = 7;
#else
  const int kCallInstructionSizeInWords = 4;
#endif
  return kCallInstructionSizeInWords * Assembler::kInstrSize;
}


27 28 29 30 31 32
void Deoptimizer::EnsureRelocSpaceForLazyDeoptimization(Handle<Code> code) {
  // Empty because there is no need for relocation information for the code
  // patching in Deoptimizer::PatchCodeForDeoptimization below.
}


33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
void Deoptimizer::PatchCodeForDeoptimization(Isolate* isolate, Code* code) {
  Address code_start_address = code->instruction_start();

  // Invalidate the relocation information, as it will become invalid by the
  // code patching below, and is not needed any more.
  code->InvalidateRelocation();

  if (FLAG_zap_code_space) {
    // Fail hard and early if we enter this code object again.
    byte* pointer = code->FindCodeAgeSequence();
    if (pointer != NULL) {
      pointer += kNoCodeAgeSequenceLength;
    } else {
      pointer = code->instruction_start();
    }
48
    CodePatcher patcher(isolate, pointer, 1);
49 50 51 52 53 54
    patcher.masm()->bkpt(0);

    DeoptimizationInputData* data =
        DeoptimizationInputData::cast(code->deoptimization_data());
    int osr_offset = data->OsrPcOffset()->value();
    if (osr_offset > 0) {
55 56
      CodePatcher osr_patcher(isolate, code->instruction_start() + osr_offset,
                              1);
57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
      osr_patcher.masm()->bkpt(0);
    }
  }

  DeoptimizationInputData* deopt_data =
      DeoptimizationInputData::cast(code->deoptimization_data());
#ifdef DEBUG
  Address prev_call_address = NULL;
#endif
  // For each LLazyBailout instruction insert a call to the corresponding
  // deoptimization entry.
  for (int i = 0; i < deopt_data->DeoptCount(); i++) {
    if (deopt_data->Pc(i)->value() == -1) continue;
    Address call_address = code_start_address + deopt_data->Pc(i)->value();
    Address deopt_entry = GetDeoptimizationEntry(isolate, i, LAZY);
    // We need calls to have a predictable size in the unoptimized code, but
    // this is optimized code, so we don't have to have a predictable size.
    int call_size_in_bytes = MacroAssembler::CallSizeNotPredictableCodeSize(
        deopt_entry, kRelocInfo_NONEPTR);
    int call_size_in_words = call_size_in_bytes / Assembler::kInstrSize;
    DCHECK(call_size_in_bytes % Assembler::kInstrSize == 0);
    DCHECK(call_size_in_bytes <= patch_size());
79
    CodePatcher patcher(isolate, call_address, call_size_in_words);
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102
    patcher.masm()->Call(deopt_entry, kRelocInfo_NONEPTR);
    DCHECK(prev_call_address == NULL ||
           call_address >= prev_call_address + patch_size());
    DCHECK(call_address + patch_size() <= code->instruction_end());
#ifdef DEBUG
    prev_call_address = call_address;
#endif
  }
}


void Deoptimizer::SetPlatformCompiledStubRegisters(
    FrameDescription* output_frame, CodeStubDescriptor* descriptor) {
  ApiFunction function(descriptor->deoptimization_handler());
  ExternalReference xref(&function, ExternalReference::BUILTIN_CALL, isolate_);
  intptr_t handler = reinterpret_cast<intptr_t>(xref.address());
  int params = descriptor->GetHandlerParameterCount();
  output_frame->SetRegister(r3.code(), params);
  output_frame->SetRegister(r4.code(), handler);
}


void Deoptimizer::CopyDoubleRegisters(FrameDescription* output_frame) {
103
  for (int i = 0; i < DoubleRegister::kNumRegisters; ++i) {
104 105 106 107 108 109 110 111 112
    double double_value = input_->GetDoubleRegister(i);
    output_frame->SetDoubleRegister(i, double_value);
  }
}

#define __ masm()->

// This code tries to be close to ia32 code so that any changes can be
// easily ported.
113
void Deoptimizer::TableEntryGenerator::Generate() {
114 115 116 117 118 119 120 121 122
  GeneratePrologue();

  // Unlike on ARM we don't save all the registers, just the useful ones.
  // For the rest, there are gaps on the stack, so the offsets remain the same.
  const int kNumberOfRegisters = Register::kNumRegisters;

  RegList restored_regs = kJSCallerSaved | kCalleeSaved;
  RegList saved_regs = restored_regs | sp.bit();

123
  const int kDoubleRegsSize = kDoubleSize * DoubleRegister::kNumRegisters;
124

125
  // Save all double registers before messing with them.
126
  __ subi(sp, sp, Operand(kDoubleRegsSize));
127 128
  const RegisterConfiguration* config =
      RegisterConfiguration::ArchDefault(RegisterConfiguration::CRANKSHAFT);
129 130 131 132 133
  for (int i = 0; i < config->num_allocatable_double_registers(); ++i) {
    int code = config->GetAllocatableDoubleCode(i);
    const DoubleRegister dreg = DoubleRegister::from_code(code);
    int offset = code * kDoubleSize;
    __ stfd(dreg, MemOperand(sp, offset));
134 135 136 137 138 139 140 141 142 143 144
  }

  // Push saved_regs (needed to populate FrameDescription::registers_).
  // Leave gaps for other registers.
  __ subi(sp, sp, Operand(kNumberOfRegisters * kPointerSize));
  for (int16_t i = kNumberOfRegisters - 1; i >= 0; i--) {
    if ((saved_regs & (1 << i)) != 0) {
      __ StoreP(ToRegister(i), MemOperand(sp, kPointerSize * i));
    }
  }

145 146 147
  __ mov(ip, Operand(ExternalReference(Isolate::kCEntryFPAddress, isolate())));
  __ StoreP(fp, MemOperand(ip));

148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
  const int kSavedRegistersAreaSize =
      (kNumberOfRegisters * kPointerSize) + kDoubleRegsSize;

  // Get the bailout id from the stack.
  __ LoadP(r5, MemOperand(sp, kSavedRegistersAreaSize));

  // Get the address of the location in the code object (r6) (return
  // address for lazy deoptimization) and compute the fp-to-sp delta in
  // register r7.
  __ mflr(r6);
  // Correct one word for bailout id.
  __ addi(r7, sp, Operand(kSavedRegistersAreaSize + (1 * kPointerSize)));
  __ sub(r7, fp, r7);

  // Allocate a new deoptimizer object.
  // Pass six arguments in r3 to r8.
  __ PrepareCallCFunction(6, r8);
165 166 167 168
  __ li(r3, Operand::Zero());
  Label context_check;
  __ LoadP(r4, MemOperand(fp, CommonFrameConstants::kContextOrFrameTypeOffset));
  __ JumpIfSmi(r4, &context_check);
169
  __ LoadP(r3, MemOperand(fp, JavaScriptFrameConstants::kFunctionOffset));
170
  __ bind(&context_check);
171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194
  __ li(r4, Operand(type()));  // bailout type,
  // r5: bailout id already loaded.
  // r6: code address or 0 already loaded.
  // r7: Fp-to-sp delta.
  __ mov(r8, Operand(ExternalReference::isolate_address(isolate())));
  // Call Deoptimizer::New().
  {
    AllowExternalCallThatCantCauseGC scope(masm());
    __ CallCFunction(ExternalReference::new_deoptimizer_function(isolate()), 6);
  }

  // Preserve "deoptimizer" object in register r3 and get the input
  // frame descriptor pointer to r4 (deoptimizer->input_);
  __ LoadP(r4, MemOperand(r3, Deoptimizer::input_offset()));

  // Copy core registers into FrameDescription::registers_[kNumRegisters].
  DCHECK(Register::kNumRegisters == kNumberOfRegisters);
  for (int i = 0; i < kNumberOfRegisters; i++) {
    int offset = (i * kPointerSize) + FrameDescription::registers_offset();
    __ LoadP(r5, MemOperand(sp, i * kPointerSize));
    __ StoreP(r5, MemOperand(r4, offset));
  }

  int double_regs_offset = FrameDescription::double_registers_offset();
195 196 197 198 199 200
  // Copy double registers to
  // double_registers_[DoubleRegister::kNumRegisters]
  for (int i = 0; i < config->num_allocatable_double_registers(); ++i) {
    int code = config->GetAllocatableDoubleCode(i);
    int dst_offset = code * kDoubleSize + double_regs_offset;
    int src_offset = code * kDoubleSize + kNumberOfRegisters * kPointerSize;
201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239
    __ lfd(d0, MemOperand(sp, src_offset));
    __ stfd(d0, MemOperand(r4, dst_offset));
  }

  // Remove the bailout id and the saved registers from the stack.
  __ addi(sp, sp, Operand(kSavedRegistersAreaSize + (1 * kPointerSize)));

  // Compute a pointer to the unwinding limit in register r5; that is
  // the first stack slot not part of the input frame.
  __ LoadP(r5, MemOperand(r4, FrameDescription::frame_size_offset()));
  __ add(r5, r5, sp);

  // Unwind the stack down to - but not including - the unwinding
  // limit and copy the contents of the activation frame to the input
  // frame description.
  __ addi(r6, r4, Operand(FrameDescription::frame_content_offset()));
  Label pop_loop;
  Label pop_loop_header;
  __ b(&pop_loop_header);
  __ bind(&pop_loop);
  __ pop(r7);
  __ StoreP(r7, MemOperand(r6, 0));
  __ addi(r6, r6, Operand(kPointerSize));
  __ bind(&pop_loop_header);
  __ cmp(r5, sp);
  __ bne(&pop_loop);

  // Compute the output frame in the deoptimizer.
  __ push(r3);  // Preserve deoptimizer object across call.
  // r3: deoptimizer object; r4: scratch.
  __ PrepareCallCFunction(1, r4);
  // Call Deoptimizer::ComputeOutputFrames().
  {
    AllowExternalCallThatCantCauseGC scope(masm());
    __ CallCFunction(
        ExternalReference::compute_output_frames_function(isolate()), 1);
  }
  __ pop(r3);  // Restore deoptimizer object (class Deoptimizer).

240 241
  __ LoadP(sp, MemOperand(r3, Deoptimizer::caller_frame_top_offset()));

242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
  // Replace the current (input) frame with the output frames.
  Label outer_push_loop, inner_push_loop, outer_loop_header, inner_loop_header;
  // Outer loop state: r7 = current "FrameDescription** output_",
  // r4 = one past the last FrameDescription**.
  __ lwz(r4, MemOperand(r3, Deoptimizer::output_count_offset()));
  __ LoadP(r7, MemOperand(r3, Deoptimizer::output_offset()));  // r7 is output_.
  __ ShiftLeftImm(r4, r4, Operand(kPointerSizeLog2));
  __ add(r4, r7, r4);
  __ b(&outer_loop_header);

  __ bind(&outer_push_loop);
  // Inner loop state: r5 = current FrameDescription*, r6 = loop index.
  __ LoadP(r5, MemOperand(r7, 0));  // output_[ix]
  __ LoadP(r6, MemOperand(r5, FrameDescription::frame_size_offset()));
  __ b(&inner_loop_header);

  __ bind(&inner_push_loop);
  __ addi(r6, r6, Operand(-sizeof(intptr_t)));
  __ add(r9, r5, r6);
  __ LoadP(r9, MemOperand(r9, FrameDescription::frame_content_offset()));
  __ push(r9);

  __ bind(&inner_loop_header);
  __ cmpi(r6, Operand::Zero());
  __ bne(&inner_push_loop);  // test for gt?

  __ addi(r7, r7, Operand(kPointerSize));
  __ bind(&outer_loop_header);
  __ cmp(r7, r4);
  __ blt(&outer_push_loop);

  __ LoadP(r4, MemOperand(r3, Deoptimizer::input_offset()));
274 275 276 277
  for (int i = 0; i < config->num_allocatable_double_registers(); ++i) {
    int code = config->GetAllocatableDoubleCode(i);
    const DoubleRegister dreg = DoubleRegister::from_code(code);
    int src_offset = code * kDoubleSize + double_regs_offset;
278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
    __ lfd(dreg, MemOperand(r4, src_offset));
  }

  // Push state, pc, and continuation from the last output frame.
  __ LoadP(r9, MemOperand(r5, FrameDescription::state_offset()));
  __ push(r9);
  __ LoadP(r9, MemOperand(r5, FrameDescription::pc_offset()));
  __ push(r9);
  __ LoadP(r9, MemOperand(r5, FrameDescription::continuation_offset()));
  __ push(r9);

  // Restore the registers from the last output frame.
  DCHECK(!(ip.bit() & restored_regs));
  __ mr(ip, r5);
  for (int i = kNumberOfRegisters - 1; i >= 0; i--) {
    int offset = (i * kPointerSize) + FrameDescription::registers_offset();
    if ((restored_regs & (1 << i)) != 0) {
      __ LoadP(ToRegister(i), MemOperand(ip, offset));
    }
  }

  __ InitializeRootRegister();

  __ pop(ip);  // get continuation, leave pc on stack
  __ pop(r0);
  __ mtlr(r0);
  __ Jump(ip);
  __ stop("Unreachable.");
}


void Deoptimizer::TableEntryGenerator::GeneratePrologue() {
  Assembler::BlockTrampolinePoolScope block_trampoline_pool(masm());

  // Create a sequence of deoptimization entries.
  // Note that registers are still live when jumping to an entry.
  Label done;
  for (int i = 0; i < count(); i++) {
    int start = masm()->pc_offset();
    USE(start);
    __ li(ip, Operand(i));
    __ b(&done);
    DCHECK(masm()->pc_offset() - start == table_entry_size_);
  }
  __ bind(&done);
  __ push(ip);
}


void FrameDescription::SetCallerPc(unsigned offset, intptr_t value) {
  SetFrameSlot(offset, value);
}


void FrameDescription::SetCallerFp(unsigned offset, intptr_t value) {
  SetFrameSlot(offset, value);
}


void FrameDescription::SetCallerConstantPool(unsigned offset, intptr_t value) {
338 339
  DCHECK(FLAG_enable_embedded_constant_pool);
  SetFrameSlot(offset, value);
340 341 342 343
}


#undef __
344 345
}  // namespace internal
}  // namespace v8