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
|
// Copyright 2018 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/wasm/jump-table-assembler.h"
#include "src/assembler-inl.h"
#include "src/macro-assembler-inl.h"
namespace v8 {
namespace internal {
namespace wasm {
// The implementation is compact enough to implement it inline here. If it gets
// much bigger, we might want to split it in a separate file per architecture.
#if V8_TARGET_ARCH_X64
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
// TODO(clemensh): Try more efficient sequences.
// Alternative 1:
// [header]: mov r10, [lazy_compile_target]
// jmp r10
// [slot 0]: push [0]
// jmp [header] // pc-relative --> slot size: 10 bytes
//
// Alternative 2:
// [header]: lea r10, [rip - [header]]
// shr r10, 3 // compute index from offset
// push r10
// mov r10, [lazy_compile_target]
// jmp r10
// [slot 0]: call [header]
// ret // -> slot size: 5 bytes
// Use a push, because mov to an extended register takes 6 bytes.
pushq(Immediate(func_index)); // max 5 bytes
movq(kScratchRegister, uint64_t{lazy_compile_target}); // max 10 bytes
jmp(kScratchRegister); // 3 bytes
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
movq(kScratchRegister, static_cast<uint64_t>(target));
jmp(kScratchRegister);
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
Nop(bytes);
}
#elif V8_TARGET_ARCH_IA32
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
mov(edi, func_index); // 5 bytes
jmp(lazy_compile_target, RelocInfo::NONE); // 5 bytes
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
jmp(target, RelocInfo::NONE);
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
Nop(bytes);
}
#elif V8_TARGET_ARCH_ARM
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
// Load function index to r4.
// This generates [movw, movt] on ARMv7 and later, [ldr, constant pool marker,
// constant] on ARMv6.
Move32BitImmediate(r4, Operand(func_index));
// EmitJumpSlot emits either [b], [movw, movt, mov] (ARMv7+), or [ldr,
// constant].
// In total, this is <=5 instructions on all architectures.
// TODO(arm): Optimize this for code size; lazy compile is not performance
// critical, as it's only executed once per function.
EmitJumpSlot(lazy_compile_target);
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
int offset =
target - reinterpret_cast<Address>(pc_) - Instruction::kPcLoadDelta;
DCHECK_EQ(0, offset % kInstrSize);
// If the offset is within 64 MB, emit a direct jump. Otherwise jump
// indirectly.
if (is_int26(offset)) {
b(offset); // 1 instr
} else {
// {Move32BitImmediate} emits either [movw, movt, mov] or [ldr, constant].
Move32BitImmediate(pc, Operand(target));
}
CheckConstPool(true, false); // force emit of const pool
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
DCHECK_EQ(0, bytes % kInstrSize);
for (; bytes > 0; bytes -= kInstrSize) {
nop();
}
}
#elif V8_TARGET_ARCH_ARM64
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
Mov(w8, func_index); // max. 2 instr
Jump(lazy_compile_target, RelocInfo::NONE); // 1 instr
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
Jump(target, RelocInfo::NONE);
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
DCHECK_EQ(0, bytes % kInstructionSize);
for (; bytes > 0; bytes -= kInstructionSize) {
nop();
}
}
#elif V8_TARGET_ARCH_S390
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
// Load function index to r7. 6 bytes
lgfi(r7, Operand(func_index));
// Jump to {lazy_compile_target}. 6 bytes or 12 bytes
mov(r1, Operand(lazy_compile_target));
b(r1); // 2 bytes
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
mov(r1, Operand(target));
b(r1);
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
DCHECK_EQ(0, bytes % 2);
for (; bytes > 0; bytes -= 2) {
nop(0);
}
}
#elif V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
li(t0, func_index); // max. 2 instr
// Jump produces max. 4 instructions for 32-bit platform
// and max. 6 instructions for 64-bit platform.
Jump(lazy_compile_target, RelocInfo::NONE);
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
Jump(target, RelocInfo::NONE);
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
DCHECK_EQ(0, bytes % kInstrSize);
for (; bytes > 0; bytes -= kInstrSize) {
nop();
}
}
#elif V8_TARGET_ARCH_PPC
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
// Load function index to r8. max 5 instrs
mov(r15, Operand(func_index));
// Jump to {lazy_compile_target}. max 5 instrs
mov(r0, Operand(lazy_compile_target));
mtctr(r0);
bctr();
}
void JumpTableAssembler::EmitJumpSlot(Address target) {
mov(r0, Operand(target));
mtctr(r0);
bctr();
}
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
DCHECK_EQ(0, bytes % 4);
for (; bytes > 0; bytes -= 4) {
nop(0);
}
}
#else
void JumpTableAssembler::EmitLazyCompileJumpSlot(uint32_t func_index,
Address lazy_compile_target) {
UNIMPLEMENTED();
}
void JumpTableAssembler::EmitJumpSlot(Address target) { UNIMPLEMENTED(); }
void JumpTableAssembler::NopBytes(int bytes) {
DCHECK_LE(0, bytes);
UNIMPLEMENTED();
}
#endif
} // namespace wasm
} // namespace internal
} // namespace v8
|
