File: deoptimizer-s390.cc

package info (click to toggle)
nodejs 4.8.2~dfsg-1
  • links: PTS, VCS
  • area: main
  • in suites: stretch
  • size: 62,476 kB
  • ctags: 111,183
  • sloc: cpp: 661,544; ansic: 31,406; python: 23,073; makefile: 1,418; sh: 1,384; perl: 255; lisp: 222; ruby: 76; xml: 50
file content (358 lines) | stat: -rw-r--r-- 12,672 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
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
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
182
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
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
240
241
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
274
275
276
277
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
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
// Copyright 2015 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/v8.h"

#include "src/codegen.h"
#include "src/deoptimizer.h"
#include "src/full-codegen.h"
#include "src/safepoint-table.h"

namespace v8 {
namespace internal {

// LAY + LGHI/LHI + BRCL
const int Deoptimizer::table_entry_size_ = 16;

int Deoptimizer::patch_size() {
#if V8_TARGET_ARCH_S390X
  const int kCallInstructionSize = 16;
#else
  const int kCallInstructionSize = 10;
#endif
  return kCallInstructionSize;
}


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


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();
    }
    CodePatcher patcher(pointer, 2);
    patcher.masm()->bkpt(0);

    DeoptimizationInputData* data =
        DeoptimizationInputData::cast(code->deoptimization_data());
    int osr_offset = data->OsrPcOffset()->value();
    if (osr_offset > 0) {
      CodePatcher osr_patcher(code->instruction_start() + osr_offset, 2);
      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);
    DCHECK(call_size_in_bytes <= patch_size());
    CodePatcher patcher(call_address, call_size_in_bytes);  // FIXME: 2ND ARG
    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::FillInputFrame(Address tos, JavaScriptFrame* frame) {
  // Set the register values. The values are not important as there are no
  // callee saved registers in JavaScript frames, so all registers are
  // spilled. Registers fp and sp are set to the correct values though.
  // We ensure the values are Smis to avoid confusing the garbage
  // collector in the event that any values are retreived and stored
  // elsewhere.

  for (int i = 0; i < Register::kNumRegisters; i++) {
    input_->SetRegister(i, reinterpret_cast<intptr_t>(Smi::FromInt(i)));
  }
  input_->SetRegister(sp.code(), reinterpret_cast<intptr_t>(frame->sp()));
  input_->SetRegister(fp.code(), reinterpret_cast<intptr_t>(frame->fp()));
  for (int i = 0; i < DoubleRegister::NumAllocatableRegisters(); i++) {
    input_->SetDoubleRegister(i, 0.0);
  }

  // Fill the frame content from the actual data on the frame.
  for (unsigned i = 0; i < input_->GetFrameSize(); i += kPointerSize) {
    input_->SetFrameSlot(
        i, reinterpret_cast<intptr_t>(Memory::Address_at(tos + i)));
  }
}


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(r2.code(), params);
  output_frame->SetRegister(r3.code(), handler);
}


void Deoptimizer::CopyDoubleRegisters(FrameDescription* output_frame) {
  for (int i = 0; i < DoubleRegister::kMaxNumRegisters; ++i) {
    double double_value = input_->GetDoubleRegister(i);
    output_frame->SetDoubleRegister(i, double_value);
  }
}


bool Deoptimizer::HasAlignmentPadding(JSFunction* function) {
  // There is no dynamic alignment padding on S390 in the input frame.
  return false;
}


#define __ masm()->

// This code tries to be close to ia32 code so that any changes can be
// easily ported.
void Deoptimizer::TableEntryGenerator::Generate() {
  GeneratePrologue();

  // Save all the registers onto the stack
  const int kNumberOfRegisters = Register::kNumRegisters;

  RegList restored_regs = kJSCallerSaved | kCalleeSaved;

  const int kDoubleRegsSize =
      kDoubleSize * DoubleRegister::kMaxNumAllocatableRegisters;

  // Save all FPU registers before messing with them.
  __ lay(sp, MemOperand(sp, -kDoubleRegsSize));
  for (int i = 0; i < DoubleRegister::kMaxNumAllocatableRegisters; ++i) {
    DoubleRegister fpu_reg = DoubleRegister::FromAllocationIndex(i);
    int offset = i * kDoubleSize;
    __ StoreF(fpu_reg, MemOperand(sp, offset));
  }

  // Push all GPRs onto the stack
  __ lay(sp, MemOperand(sp, -kNumberOfRegisters * kPointerSize));
  __ StoreMultipleP(r0, sp, MemOperand(sp));   // Save all 16 registers

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

  const int kSavedRegistersAreaSize =
      (kNumberOfRegisters * kPointerSize) + kDoubleRegsSize;

  // Get the bailout id from the stack.
  __ LoadP(r4, MemOperand(sp, kSavedRegistersAreaSize));

  // Cleanse the Return address for 31-bit
  __ CleanseP(r14);

  // Get the address of the location in the code object (r5)(return
  // address for lazy deoptimization) and compute the fp-to-sp delta in
  // register r6.
  __ LoadRR(r5, r14);
  __ la(r6, MemOperand(sp, kSavedRegistersAreaSize + (1 * kPointerSize)));
  __ SubP(r6, fp, r6);

  // Allocate a new deoptimizer object.
  // Pass six arguments in r2 to r7.
  __ PrepareCallCFunction(6, r7);
  __ LoadP(r2, MemOperand(fp, JavaScriptFrameConstants::kFunctionOffset));
  __ LoadImmP(r3, Operand(type()));  // bailout type,
  // r4: bailout id already loaded.
  // r5: code address or 0 already loaded.
  // r6: Fp-to-sp delta.
  // Parm6: isolate is passed on the stack.
  __ mov(r7, Operand(ExternalReference::isolate_address(isolate())));
  __ StoreP(r7, MemOperand(sp, kStackFrameExtraParamSlot * kPointerSize));

  // Call Deoptimizer::New().
  {
    AllowExternalCallThatCantCauseGC scope(masm());
    __ CallCFunction(ExternalReference::new_deoptimizer_function(isolate()), 6);
  }

  // Preserve "deoptimizer" object in register r2 and get the input
  // frame descriptor pointer to r3 (deoptimizer->input_);
  __ LoadP(r3, MemOperand(r2, Deoptimizer::input_offset()));

  // Copy core registers into FrameDescription::registers_[kNumRegisters].
  DCHECK(Register::kNumRegisters == kNumberOfRegisters);
  __ mvc(MemOperand(r3, FrameDescription::registers_offset()), MemOperand(sp),
         kNumberOfRegisters * kPointerSize);

  int double_regs_offset = FrameDescription::double_registers_offset();
  // Copy VFP registers to
  // double_registers_[DoubleRegister::kNumAllocatableRegisters]
  __ mvc(MemOperand(r3, double_regs_offset),
         MemOperand(sp, kNumberOfRegisters * kPointerSize),
         DoubleRegister::NumAllocatableRegisters() * kDoubleSize);

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

  // Compute a pointer to the unwinding limit in register r4; that is
  // the first stack slot not part of the input frame.
  __ LoadP(r4, MemOperand(r3, FrameDescription::frame_size_offset()));
  __ AddP(r4, 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.
  __ la(r5, MemOperand(r3, FrameDescription::frame_content_offset()));
  Label pop_loop;
  Label pop_loop_header;
  __ b(&pop_loop_header, Label::kNear);
  __ bind(&pop_loop);
  __ pop(r6);
  __ StoreP(r6, MemOperand(r5, 0));
  __ la(r5, MemOperand(r5, kPointerSize));
  __ bind(&pop_loop_header);
  __ CmpP(r4, sp);
  __ bne(&pop_loop);

  // Compute the output frame in the deoptimizer.
  __ push(r2);  // Preserve deoptimizer object across call.
  // r2: deoptimizer object; r3: scratch.
  __ PrepareCallCFunction(1, r3);
  // Call Deoptimizer::ComputeOutputFrames().
  {
    AllowExternalCallThatCantCauseGC scope(masm());
    __ CallCFunction(
        ExternalReference::compute_output_frames_function(isolate()), 1);
  }
  __ pop(r2);  // Restore deoptimizer object (class Deoptimizer).

  // 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: r6 = current "FrameDescription** output_",
  // r3 = one past the last FrameDescription**.
  __ LoadlW(r3, MemOperand(r2, Deoptimizer::output_count_offset()));
  __ LoadP(r6, MemOperand(r2, Deoptimizer::output_offset()));  // r6 is output_.
  __ ShiftLeftP(r3, r3, Operand(kPointerSizeLog2));
  __ AddP(r3, r6, r3);
  __ b(&outer_loop_header, Label::kNear);

  __ bind(&outer_push_loop);
  // Inner loop state: r4 = current FrameDescription*, r5 = loop index.
  __ LoadP(r4, MemOperand(r6, 0));  // output_[ix]
  __ LoadP(r5, MemOperand(r4, FrameDescription::frame_size_offset()));
  __ b(&inner_loop_header, Label::kNear);

  __ bind(&inner_push_loop);
  __ AddP(r5, Operand(-sizeof(intptr_t)));
  __ AddP(r8, r4, r5);
  __ LoadP(r8, MemOperand(r8, FrameDescription::frame_content_offset()));
  __ push(r8);

  __ bind(&inner_loop_header);
  __ CmpP(r5, Operand::Zero());
  __ bne(&inner_push_loop);  // test for gt?

  __ AddP(r6, r6, Operand(kPointerSize));
  __ bind(&outer_loop_header);
  __ CmpP(r6, r3);
  __ blt(&outer_push_loop);

  __ LoadP(r3, MemOperand(r2, Deoptimizer::input_offset()));
  for (int i = 0; i < DoubleRegister::kMaxNumAllocatableRegisters; ++i) {
    const DoubleRegister dreg = DoubleRegister::FromAllocationIndex(i);
    int src_offset = i * kDoubleSize + double_regs_offset;
    __ ld(dreg, MemOperand(r3, src_offset));
  }

  // Push state, pc, and continuation from the last output frame.
  __ LoadP(r8, MemOperand(r4, FrameDescription::state_offset()));
  __ push(r8);
  __ LoadP(r8, MemOperand(r4, FrameDescription::pc_offset()));
  __ push(r8);
  __ LoadP(r8, MemOperand(r4, FrameDescription::continuation_offset()));
  __ push(r8);

  // Restore the registers from the last output frame.
  DCHECK(!(ip.bit() & restored_regs));
  __ LoadRR(ip, r4);
  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(r14);
  __ Jump(ip);
  __ stop("Unreachable.");
}


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

  // Create a sequence of deoptimization entries. Note that any
  // registers may be still live.
  Label done;
  for (int i = 0; i < count(); i++) {
    int start = masm()->pc_offset();
    USE(start);
    __ lay(sp, MemOperand(sp, -kPointerSize));
    __ LoadImmP(ip, Operand(i));
    __ b(&done);
    int end = masm()->pc_offset();
    USE(end);
    DCHECK(masm()->pc_offset() - start == table_entry_size_);
  }
  __ bind(&done);
  __ StoreP(ip, MemOperand(sp));
}


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) {
  // No out-of-line constant pool support.
  UNREACHABLE();
}


#undef __
}
}  // namespace v8::internal