1 /*
2 * Copyright (C) 2012 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include "codegen_x86.h"
18 #include "dex/quick/mir_to_lir-inl.h"
19 #include "x86_lir.h"
20
21 namespace art {
22
23 #define MAX_ASSEMBLER_RETRIES 50
24
25 const X86EncodingMap X86Mir2Lir::EncodingMap[kX86Last] = {
26 { kX8632BitData, kData, IS_UNARY_OP, { 0, 0, 0x00, 0, 0, 0, 0, 4 }, "data", "0x!0d" },
27 { kX86Bkpt, kNullary, NO_OPERAND | IS_BRANCH, { 0, 0, 0xCC, 0, 0, 0, 0, 0 }, "int 3", "" },
28 { kX86Nop, kNop, IS_UNARY_OP, { 0, 0, 0x90, 0, 0, 0, 0, 0 }, "nop", "" },
29
30 #define ENCODING_MAP(opname, mem_use, reg_def, uses_ccodes, \
31 rm8_r8, rm32_r32, \
32 r8_rm8, r32_rm32, \
33 ax8_i8, ax32_i32, \
34 rm8_i8, rm8_i8_modrm, \
35 rm32_i32, rm32_i32_modrm, \
36 rm32_i8, rm32_i8_modrm) \
37 { kX86 ## opname ## 8MR, kMemReg, mem_use | IS_TERTIARY_OP | REG_USE02 | SETS_CCODES | uses_ccodes, { 0, 0, rm8_r8, 0, 0, 0, 0, 0 }, #opname "8MR", "[!0r+!1d],!2r" }, \
38 { kX86 ## opname ## 8AR, kArrayReg, mem_use | IS_QUIN_OP | REG_USE014 | SETS_CCODES | uses_ccodes, { 0, 0, rm8_r8, 0, 0, 0, 0, 0 }, #opname "8AR", "[!0r+!1r<<!2d+!3d],!4r" }, \
39 { kX86 ## opname ## 8TR, kThreadReg, mem_use | IS_BINARY_OP | REG_USE1 | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, rm8_r8, 0, 0, 0, 0, 0 }, #opname "8TR", "fs:[!0d],!1r" }, \
40 { kX86 ## opname ## 8RR, kRegReg, IS_BINARY_OP | reg_def | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, r8_rm8, 0, 0, 0, 0, 0 }, #opname "8RR", "!0r,!1r" }, \
41 { kX86 ## opname ## 8RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | reg_def | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, r8_rm8, 0, 0, 0, 0, 0 }, #opname "8RM", "!0r,[!1r+!2d]" }, \
42 { kX86 ## opname ## 8RA, kRegArray, IS_LOAD | IS_QUIN_OP | reg_def | REG_USE012 | SETS_CCODES | uses_ccodes, { 0, 0, r8_rm8, 0, 0, 0, 0, 0 }, #opname "8RA", "!0r,[!1r+!2r<<!3d+!4d]" }, \
43 { kX86 ## opname ## 8RT, kRegThread, IS_LOAD | IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, r8_rm8, 0, 0, 0, 0, 0 }, #opname "8RT", "!0r,fs:[!1d]" }, \
44 { kX86 ## opname ## 8RI, kRegImm, IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { 0, 0, rm8_i8, 0, 0, rm8_i8_modrm, ax8_i8, 1 }, #opname "8RI", "!0r,!1d" }, \
45 { kX86 ## opname ## 8MI, kMemImm, mem_use | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES | uses_ccodes, { 0, 0, rm8_i8, 0, 0, rm8_i8_modrm, 0, 1 }, #opname "8MI", "[!0r+!1d],!2d" }, \
46 { kX86 ## opname ## 8AI, kArrayImm, mem_use | IS_QUIN_OP | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, rm8_i8, 0, 0, rm8_i8_modrm, 0, 1 }, #opname "8AI", "[!0r+!1r<<!2d+!3d],!4d" }, \
47 { kX86 ## opname ## 8TI, kThreadImm, mem_use | IS_BINARY_OP | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, rm8_i8, 0, 0, rm8_i8_modrm, 0, 1 }, #opname "8TI", "fs:[!0d],!1d" }, \
48 \
49 { kX86 ## opname ## 16MR, kMemReg, mem_use | IS_TERTIARY_OP | REG_USE02 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_r32, 0, 0, 0, 0, 0 }, #opname "16MR", "[!0r+!1d],!2r" }, \
50 { kX86 ## opname ## 16AR, kArrayReg, mem_use | IS_QUIN_OP | REG_USE014 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_r32, 0, 0, 0, 0, 0 }, #opname "16AR", "[!0r+!1r<<!2d+!3d],!4r" }, \
51 { kX86 ## opname ## 16TR, kThreadReg, mem_use | IS_BINARY_OP | REG_USE1 | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0x66, rm32_r32, 0, 0, 0, 0, 0 }, #opname "16TR", "fs:[!0d],!1r" }, \
52 { kX86 ## opname ## 16RR, kRegReg, IS_BINARY_OP | reg_def | REG_USE01 | SETS_CCODES | uses_ccodes, { 0x66, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "16RR", "!0r,!1r" }, \
53 { kX86 ## opname ## 16RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | reg_def | REG_USE01 | SETS_CCODES | uses_ccodes, { 0x66, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "16RM", "!0r,[!1r+!2d]" }, \
54 { kX86 ## opname ## 16RA, kRegArray, IS_LOAD | IS_QUIN_OP | reg_def | REG_USE012 | SETS_CCODES | uses_ccodes, { 0x66, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "16RA", "!0r,[!1r+!2r<<!3d+!4d]" }, \
55 { kX86 ## opname ## 16RT, kRegThread, IS_LOAD | IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0x66, r32_rm32, 0, 0, 0, 0, 0 }, #opname "16RT", "!0r,fs:[!1d]" }, \
56 { kX86 ## opname ## 16RI, kRegImm, IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_i32, 0, 0, rm32_i32_modrm, ax32_i32, 2 }, #opname "16RI", "!0r,!1d" }, \
57 { kX86 ## opname ## 16MI, kMemImm, mem_use | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_i32, 0, 0, rm32_i32_modrm, 0, 2 }, #opname "16MI", "[!0r+!1d],!2d" }, \
58 { kX86 ## opname ## 16AI, kArrayImm, mem_use | IS_QUIN_OP | REG_USE01 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_i32, 0, 0, rm32_i32_modrm, 0, 2 }, #opname "16AI", "[!0r+!1r<<!2d+!3d],!4d" }, \
59 { kX86 ## opname ## 16TI, kThreadImm, mem_use | IS_BINARY_OP | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0x66, rm32_i32, 0, 0, rm32_i32_modrm, 0, 2 }, #opname "16TI", "fs:[!0d],!1d" }, \
60 { kX86 ## opname ## 16RI8, kRegImm, IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "16RI8", "!0r,!1d" }, \
61 { kX86 ## opname ## 16MI8, kMemImm, mem_use | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "16MI8", "[!0r+!1d],!2d" }, \
62 { kX86 ## opname ## 16AI8, kArrayImm, mem_use | IS_QUIN_OP | REG_USE01 | SETS_CCODES | uses_ccodes, { 0x66, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "16AI8", "[!0r+!1r<<!2d+!3d],!4d" }, \
63 { kX86 ## opname ## 16TI8, kThreadImm, mem_use | IS_BINARY_OP | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0x66, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "16TI8", "fs:[!0d],!1d" }, \
64 \
65 { kX86 ## opname ## 32MR, kMemReg, mem_use | IS_TERTIARY_OP | REG_USE02 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_r32, 0, 0, 0, 0, 0 }, #opname "32MR", "[!0r+!1d],!2r" }, \
66 { kX86 ## opname ## 32AR, kArrayReg, mem_use | IS_QUIN_OP | REG_USE014 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_r32, 0, 0, 0, 0, 0 }, #opname "32AR", "[!0r+!1r<<!2d+!3d],!4r" }, \
67 { kX86 ## opname ## 32TR, kThreadReg, mem_use | IS_BINARY_OP | REG_USE1 | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, rm32_r32, 0, 0, 0, 0, 0 }, #opname "32TR", "fs:[!0d],!1r" }, \
68 { kX86 ## opname ## 32RR, kRegReg, IS_BINARY_OP | reg_def | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "32RR", "!0r,!1r" }, \
69 { kX86 ## opname ## 32RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | reg_def | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "32RM", "!0r,[!1r+!2d]" }, \
70 { kX86 ## opname ## 32RA, kRegArray, IS_LOAD | IS_QUIN_OP | reg_def | REG_USE012 | SETS_CCODES | uses_ccodes, { 0, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "32RA", "!0r,[!1r+!2r<<!3d+!4d]" }, \
71 { kX86 ## opname ## 32RT, kRegThread, IS_LOAD | IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, r32_rm32, 0, 0, 0, 0, 0 }, #opname "32RT", "!0r,fs:[!1d]" }, \
72 { kX86 ## opname ## 32RI, kRegImm, IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_i32, 0, 0, rm32_i32_modrm, ax32_i32, 4 }, #opname "32RI", "!0r,!1d" }, \
73 { kX86 ## opname ## 32MI, kMemImm, mem_use | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_i32, 0, 0, rm32_i32_modrm, 0, 4 }, #opname "32MI", "[!0r+!1d],!2d" }, \
74 { kX86 ## opname ## 32AI, kArrayImm, mem_use | IS_QUIN_OP | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_i32, 0, 0, rm32_i32_modrm, 0, 4 }, #opname "32AI", "[!0r+!1r<<!2d+!3d],!4d" }, \
75 { kX86 ## opname ## 32TI, kThreadImm, mem_use | IS_BINARY_OP | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, rm32_i32, 0, 0, rm32_i32_modrm, 0, 4 }, #opname "32TI", "fs:[!0d],!1d" }, \
76 { kX86 ## opname ## 32RI8, kRegImm, IS_BINARY_OP | reg_def | REG_USE0 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "32RI8", "!0r,!1d" }, \
77 { kX86 ## opname ## 32MI8, kMemImm, mem_use | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "32MI8", "[!0r+!1d],!2d" }, \
78 { kX86 ## opname ## 32AI8, kArrayImm, mem_use | IS_QUIN_OP | REG_USE01 | SETS_CCODES | uses_ccodes, { 0, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "32AI8", "[!0r+!1r<<!2d+!3d],!4d" }, \
79 { kX86 ## opname ## 32TI8, kThreadImm, mem_use | IS_BINARY_OP | SETS_CCODES | uses_ccodes, { THREAD_PREFIX, 0, rm32_i8, 0, 0, rm32_i8_modrm, 0, 1 }, #opname "32TI8", "fs:[!0d],!1d" }
80
81 ENCODING_MAP(Add, IS_LOAD | IS_STORE, REG_DEF0, 0,
82 0x00 /* RegMem8/Reg8 */, 0x01 /* RegMem32/Reg32 */,
83 0x02 /* Reg8/RegMem8 */, 0x03 /* Reg32/RegMem32 */,
84 0x04 /* Rax8/imm8 opcode */, 0x05 /* Rax32/imm32 */,
85 0x80, 0x0 /* RegMem8/imm8 */,
86 0x81, 0x0 /* RegMem32/imm32 */, 0x83, 0x0 /* RegMem32/imm8 */),
87 ENCODING_MAP(Or, IS_LOAD | IS_STORE, REG_DEF0, 0,
88 0x08 /* RegMem8/Reg8 */, 0x09 /* RegMem32/Reg32 */,
89 0x0A /* Reg8/RegMem8 */, 0x0B /* Reg32/RegMem32 */,
90 0x0C /* Rax8/imm8 opcode */, 0x0D /* Rax32/imm32 */,
91 0x80, 0x1 /* RegMem8/imm8 */,
92 0x81, 0x1 /* RegMem32/imm32 */, 0x83, 0x1 /* RegMem32/imm8 */),
93 ENCODING_MAP(Adc, IS_LOAD | IS_STORE, REG_DEF0, USES_CCODES,
94 0x10 /* RegMem8/Reg8 */, 0x11 /* RegMem32/Reg32 */,
95 0x12 /* Reg8/RegMem8 */, 0x13 /* Reg32/RegMem32 */,
96 0x14 /* Rax8/imm8 opcode */, 0x15 /* Rax32/imm32 */,
97 0x80, 0x2 /* RegMem8/imm8 */,
98 0x81, 0x2 /* RegMem32/imm32 */, 0x83, 0x2 /* RegMem32/imm8 */),
99 ENCODING_MAP(Sbb, IS_LOAD | IS_STORE, REG_DEF0, USES_CCODES,
100 0x18 /* RegMem8/Reg8 */, 0x19 /* RegMem32/Reg32 */,
101 0x1A /* Reg8/RegMem8 */, 0x1B /* Reg32/RegMem32 */,
102 0x1C /* Rax8/imm8 opcode */, 0x1D /* Rax32/imm32 */,
103 0x80, 0x3 /* RegMem8/imm8 */,
104 0x81, 0x3 /* RegMem32/imm32 */, 0x83, 0x3 /* RegMem32/imm8 */),
105 ENCODING_MAP(And, IS_LOAD | IS_STORE, REG_DEF0, 0,
106 0x20 /* RegMem8/Reg8 */, 0x21 /* RegMem32/Reg32 */,
107 0x22 /* Reg8/RegMem8 */, 0x23 /* Reg32/RegMem32 */,
108 0x24 /* Rax8/imm8 opcode */, 0x25 /* Rax32/imm32 */,
109 0x80, 0x4 /* RegMem8/imm8 */,
110 0x81, 0x4 /* RegMem32/imm32 */, 0x83, 0x4 /* RegMem32/imm8 */),
111 ENCODING_MAP(Sub, IS_LOAD | IS_STORE, REG_DEF0, 0,
112 0x28 /* RegMem8/Reg8 */, 0x29 /* RegMem32/Reg32 */,
113 0x2A /* Reg8/RegMem8 */, 0x2B /* Reg32/RegMem32 */,
114 0x2C /* Rax8/imm8 opcode */, 0x2D /* Rax32/imm32 */,
115 0x80, 0x5 /* RegMem8/imm8 */,
116 0x81, 0x5 /* RegMem32/imm32 */, 0x83, 0x5 /* RegMem32/imm8 */),
117 ENCODING_MAP(Xor, IS_LOAD | IS_STORE, REG_DEF0, 0,
118 0x30 /* RegMem8/Reg8 */, 0x31 /* RegMem32/Reg32 */,
119 0x32 /* Reg8/RegMem8 */, 0x33 /* Reg32/RegMem32 */,
120 0x34 /* Rax8/imm8 opcode */, 0x35 /* Rax32/imm32 */,
121 0x80, 0x6 /* RegMem8/imm8 */,
122 0x81, 0x6 /* RegMem32/imm32 */, 0x83, 0x6 /* RegMem32/imm8 */),
123 ENCODING_MAP(Cmp, IS_LOAD, 0, 0,
124 0x38 /* RegMem8/Reg8 */, 0x39 /* RegMem32/Reg32 */,
125 0x3A /* Reg8/RegMem8 */, 0x3B /* Reg32/RegMem32 */,
126 0x3C /* Rax8/imm8 opcode */, 0x3D /* Rax32/imm32 */,
127 0x80, 0x7 /* RegMem8/imm8 */,
128 0x81, 0x7 /* RegMem32/imm32 */, 0x83, 0x7 /* RegMem32/imm8 */),
129 #undef ENCODING_MAP
130
131 { kX86Imul16RRI, kRegRegImm, IS_TERTIARY_OP | REG_DEF0_USE1 | SETS_CCODES, { 0x66, 0, 0x69, 0, 0, 0, 0, 2 }, "Imul16RRI", "!0r,!1r,!2d" },
132 { kX86Imul16RMI, kRegMemImm, IS_LOAD | IS_QUAD_OP | REG_DEF0_USE1 | SETS_CCODES, { 0x66, 0, 0x69, 0, 0, 0, 0, 2 }, "Imul16RMI", "!0r,[!1r+!2d],!3d" },
133 { kX86Imul16RAI, kRegArrayImm, IS_LOAD | IS_SEXTUPLE_OP | REG_DEF0_USE12 | SETS_CCODES, { 0x66, 0, 0x69, 0, 0, 0, 0, 2 }, "Imul16RAI", "!0r,[!1r+!2r<<!3d+!4d],!5d" },
134
135 { kX86Imul32RRI, kRegRegImm, IS_TERTIARY_OP | REG_DEF0_USE1 | SETS_CCODES, { 0, 0, 0x69, 0, 0, 0, 0, 4 }, "Imul32RRI", "!0r,!1r,!2d" },
136 { kX86Imul32RMI, kRegMemImm, IS_LOAD | IS_QUAD_OP | REG_DEF0_USE1 | SETS_CCODES, { 0, 0, 0x69, 0, 0, 0, 0, 4 }, "Imul32RMI", "!0r,[!1r+!2d],!3d" },
137 { kX86Imul32RAI, kRegArrayImm, IS_LOAD | IS_SEXTUPLE_OP | REG_DEF0_USE12 | SETS_CCODES, { 0, 0, 0x69, 0, 0, 0, 0, 4 }, "Imul32RAI", "!0r,[!1r+!2r<<!3d+!4d],!5d" },
138 { kX86Imul32RRI8, kRegRegImm, IS_TERTIARY_OP | REG_DEF0_USE1 | SETS_CCODES, { 0, 0, 0x6B, 0, 0, 0, 0, 1 }, "Imul32RRI8", "!0r,!1r,!2d" },
139 { kX86Imul32RMI8, kRegMemImm, IS_LOAD | IS_QUAD_OP | REG_DEF0_USE1 | SETS_CCODES, { 0, 0, 0x6B, 0, 0, 0, 0, 1 }, "Imul32RMI8", "!0r,[!1r+!2d],!3d" },
140 { kX86Imul32RAI8, kRegArrayImm, IS_LOAD | IS_SEXTUPLE_OP | REG_DEF0_USE12 | SETS_CCODES, { 0, 0, 0x6B, 0, 0, 0, 0, 1 }, "Imul32RAI8", "!0r,[!1r+!2r<<!3d+!4d],!5d" },
141
142 { kX86Mov8MR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02, { 0, 0, 0x88, 0, 0, 0, 0, 0 }, "Mov8MR", "[!0r+!1d],!2r" },
143 { kX86Mov8AR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014, { 0, 0, 0x88, 0, 0, 0, 0, 0 }, "Mov8AR", "[!0r+!1r<<!2d+!3d],!4r" },
144 { kX86Mov8TR, kThreadReg, IS_STORE | IS_BINARY_OP | REG_USE1, { THREAD_PREFIX, 0, 0x88, 0, 0, 0, 0, 0 }, "Mov8TR", "fs:[!0d],!1r" },
145 { kX86Mov8RR, kRegReg, IS_BINARY_OP | REG_DEF0_USE1, { 0, 0, 0x8A, 0, 0, 0, 0, 0 }, "Mov8RR", "!0r,!1r" },
146 { kX86Mov8RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | REG_DEF0_USE1, { 0, 0, 0x8A, 0, 0, 0, 0, 0 }, "Mov8RM", "!0r,[!1r+!2d]" },
147 { kX86Mov8RA, kRegArray, IS_LOAD | IS_QUIN_OP | REG_DEF0_USE12, { 0, 0, 0x8A, 0, 0, 0, 0, 0 }, "Mov8RA", "!0r,[!1r+!2r<<!3d+!4d]" },
148 { kX86Mov8RT, kRegThread, IS_LOAD | IS_BINARY_OP | REG_DEF0, { THREAD_PREFIX, 0, 0x8A, 0, 0, 0, 0, 0 }, "Mov8RT", "!0r,fs:[!1d]" },
149 { kX86Mov8RI, kMovRegImm, IS_BINARY_OP | REG_DEF0, { 0, 0, 0xB0, 0, 0, 0, 0, 1 }, "Mov8RI", "!0r,!1d" },
150 { kX86Mov8MI, kMemImm, IS_STORE | IS_TERTIARY_OP | REG_USE0, { 0, 0, 0xC6, 0, 0, 0, 0, 1 }, "Mov8MI", "[!0r+!1d],!2d" },
151 { kX86Mov8AI, kArrayImm, IS_STORE | IS_QUIN_OP | REG_USE01, { 0, 0, 0xC6, 0, 0, 0, 0, 1 }, "Mov8AI", "[!0r+!1r<<!2d+!3d],!4d" },
152 { kX86Mov8TI, kThreadImm, IS_STORE | IS_BINARY_OP, { THREAD_PREFIX, 0, 0xC6, 0, 0, 0, 0, 1 }, "Mov8TI", "fs:[!0d],!1d" },
153
154 { kX86Mov16MR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02, { 0x66, 0, 0x89, 0, 0, 0, 0, 0 }, "Mov16MR", "[!0r+!1d],!2r" },
155 { kX86Mov16AR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014, { 0x66, 0, 0x89, 0, 0, 0, 0, 0 }, "Mov16AR", "[!0r+!1r<<!2d+!3d],!4r" },
156 { kX86Mov16TR, kThreadReg, IS_STORE | IS_BINARY_OP | REG_USE1, { THREAD_PREFIX, 0x66, 0x89, 0, 0, 0, 0, 0 }, "Mov16TR", "fs:[!0d],!1r" },
157 { kX86Mov16RR, kRegReg, IS_BINARY_OP | REG_DEF0_USE1, { 0x66, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov16RR", "!0r,!1r" },
158 { kX86Mov16RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | REG_DEF0_USE1, { 0x66, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov16RM", "!0r,[!1r+!2d]" },
159 { kX86Mov16RA, kRegArray, IS_LOAD | IS_QUIN_OP | REG_DEF0_USE12, { 0x66, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov16RA", "!0r,[!1r+!2r<<!3d+!4d]" },
160 { kX86Mov16RT, kRegThread, IS_LOAD | IS_BINARY_OP | REG_DEF0, { THREAD_PREFIX, 0x66, 0x8B, 0, 0, 0, 0, 0 }, "Mov16RT", "!0r,fs:[!1d]" },
161 { kX86Mov16RI, kMovRegImm, IS_BINARY_OP | REG_DEF0, { 0x66, 0, 0xB8, 0, 0, 0, 0, 2 }, "Mov16RI", "!0r,!1d" },
162 { kX86Mov16MI, kMemImm, IS_STORE | IS_TERTIARY_OP | REG_USE0, { 0x66, 0, 0xC7, 0, 0, 0, 0, 2 }, "Mov16MI", "[!0r+!1d],!2d" },
163 { kX86Mov16AI, kArrayImm, IS_STORE | IS_QUIN_OP | REG_USE01, { 0x66, 0, 0xC7, 0, 0, 0, 0, 2 }, "Mov16AI", "[!0r+!1r<<!2d+!3d],!4d" },
164 { kX86Mov16TI, kThreadImm, IS_STORE | IS_BINARY_OP, { THREAD_PREFIX, 0x66, 0xC7, 0, 0, 0, 0, 2 }, "Mov16TI", "fs:[!0d],!1d" },
165
166 { kX86Mov32MR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02, { 0, 0, 0x89, 0, 0, 0, 0, 0 }, "Mov32MR", "[!0r+!1d],!2r" },
167 { kX86Mov32AR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014, { 0, 0, 0x89, 0, 0, 0, 0, 0 }, "Mov32AR", "[!0r+!1r<<!2d+!3d],!4r" },
168 { kX86Mov32TR, kThreadReg, IS_STORE | IS_BINARY_OP | REG_USE1, { THREAD_PREFIX, 0, 0x89, 0, 0, 0, 0, 0 }, "Mov32TR", "fs:[!0d],!1r" },
169 { kX86Mov32RR, kRegReg, IS_BINARY_OP | REG_DEF0_USE1, { 0, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov32RR", "!0r,!1r" },
170 { kX86Mov32RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | REG_DEF0_USE1, { 0, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov32RM", "!0r,[!1r+!2d]" },
171 { kX86Mov32RA, kRegArray, IS_LOAD | IS_QUIN_OP | REG_DEF0_USE12, { 0, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov32RA", "!0r,[!1r+!2r<<!3d+!4d]" },
172 { kX86Mov32RT, kRegThread, IS_LOAD | IS_BINARY_OP | REG_DEF0, { THREAD_PREFIX, 0, 0x8B, 0, 0, 0, 0, 0 }, "Mov32RT", "!0r,fs:[!1d]" },
173 { kX86Mov32RI, kMovRegImm, IS_BINARY_OP | REG_DEF0, { 0, 0, 0xB8, 0, 0, 0, 0, 4 }, "Mov32RI", "!0r,!1d" },
174 { kX86Mov32MI, kMemImm, IS_STORE | IS_TERTIARY_OP | REG_USE0, { 0, 0, 0xC7, 0, 0, 0, 0, 4 }, "Mov32MI", "[!0r+!1d],!2d" },
175 { kX86Mov32AI, kArrayImm, IS_STORE | IS_QUIN_OP | REG_USE01, { 0, 0, 0xC7, 0, 0, 0, 0, 4 }, "Mov32AI", "[!0r+!1r<<!2d+!3d],!4d" },
176 { kX86Mov32TI, kThreadImm, IS_STORE | IS_BINARY_OP, { THREAD_PREFIX, 0, 0xC7, 0, 0, 0, 0, 4 }, "Mov32TI", "fs:[!0d],!1d" },
177
178 { kX86Lea32RA, kRegArray, IS_QUIN_OP | REG_DEF0_USE12, { 0, 0, 0x8D, 0, 0, 0, 0, 0 }, "Lea32RA", "!0r,[!1r+!2r<<!3d+!4d]" },
179
180 #define SHIFT_ENCODING_MAP(opname, modrm_opcode) \
181 { kX86 ## opname ## 8RI, kShiftRegImm, IS_BINARY_OP | REG_DEF0_USE0 | SETS_CCODES, { 0, 0, 0xC0, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "8RI", "!0r,!1d" }, \
182 { kX86 ## opname ## 8MI, kShiftMemImm, IS_LOAD | IS_STORE | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES, { 0, 0, 0xC0, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "8MI", "[!0r+!1d],!2d" }, \
183 { kX86 ## opname ## 8AI, kShiftArrayImm, IS_LOAD | IS_STORE | IS_QUIN_OP | REG_USE01 | SETS_CCODES, { 0, 0, 0xC0, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "8AI", "[!0r+!1r<<!2d+!3d],!4d" }, \
184 { kX86 ## opname ## 8RC, kShiftRegCl, IS_BINARY_OP | REG_DEF0_USE0 | REG_USEC | SETS_CCODES, { 0, 0, 0xD2, 0, 0, modrm_opcode, 0, 1 }, #opname "8RC", "!0r,cl" }, \
185 { kX86 ## opname ## 8MC, kShiftMemCl, IS_LOAD | IS_STORE | IS_TERTIARY_OP | REG_USE0 | REG_USEC | SETS_CCODES, { 0, 0, 0xD2, 0, 0, modrm_opcode, 0, 1 }, #opname "8MC", "[!0r+!1d],cl" }, \
186 { kX86 ## opname ## 8AC, kShiftArrayCl, IS_LOAD | IS_STORE | IS_QUIN_OP | REG_USE01 | REG_USEC | SETS_CCODES, { 0, 0, 0xD2, 0, 0, modrm_opcode, 0, 1 }, #opname "8AC", "[!0r+!1r<<!2d+!3d],cl" }, \
187 \
188 { kX86 ## opname ## 16RI, kShiftRegImm, IS_BINARY_OP | REG_DEF0_USE0 | SETS_CCODES, { 0x66, 0, 0xC1, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "16RI", "!0r,!1d" }, \
189 { kX86 ## opname ## 16MI, kShiftMemImm, IS_LOAD | IS_STORE | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES, { 0x66, 0, 0xC1, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "16MI", "[!0r+!1d],!2d" }, \
190 { kX86 ## opname ## 16AI, kShiftArrayImm, IS_LOAD | IS_STORE | IS_QUIN_OP | REG_USE01 | SETS_CCODES, { 0x66, 0, 0xC1, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "16AI", "[!0r+!1r<<!2d+!3d],!4d" }, \
191 { kX86 ## opname ## 16RC, kShiftRegCl, IS_BINARY_OP | REG_DEF0_USE0 | REG_USEC | SETS_CCODES, { 0x66, 0, 0xD3, 0, 0, modrm_opcode, 0, 1 }, #opname "16RC", "!0r,cl" }, \
192 { kX86 ## opname ## 16MC, kShiftMemCl, IS_LOAD | IS_STORE | IS_TERTIARY_OP | REG_USE0 | REG_USEC | SETS_CCODES, { 0x66, 0, 0xD3, 0, 0, modrm_opcode, 0, 1 }, #opname "16MC", "[!0r+!1d],cl" }, \
193 { kX86 ## opname ## 16AC, kShiftArrayCl, IS_LOAD | IS_STORE | IS_QUIN_OP | REG_USE01 | REG_USEC | SETS_CCODES, { 0x66, 0, 0xD3, 0, 0, modrm_opcode, 0, 1 }, #opname "16AC", "[!0r+!1r<<!2d+!3d],cl" }, \
194 \
195 { kX86 ## opname ## 32RI, kShiftRegImm, IS_BINARY_OP | REG_DEF0_USE0 | SETS_CCODES, { 0, 0, 0xC1, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "32RI", "!0r,!1d" }, \
196 { kX86 ## opname ## 32MI, kShiftMemImm, IS_LOAD | IS_STORE | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES, { 0, 0, 0xC1, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "32MI", "[!0r+!1d],!2d" }, \
197 { kX86 ## opname ## 32AI, kShiftArrayImm, IS_LOAD | IS_STORE | IS_QUIN_OP | REG_USE01 | SETS_CCODES, { 0, 0, 0xC1, 0, 0, modrm_opcode, 0xD1, 1 }, #opname "32AI", "[!0r+!1r<<!2d+!3d],!4d" }, \
198 { kX86 ## opname ## 32RC, kShiftRegCl, IS_BINARY_OP | REG_DEF0_USE0 | REG_USEC | SETS_CCODES, { 0, 0, 0xD3, 0, 0, modrm_opcode, 0, 0 }, #opname "32RC", "!0r,cl" }, \
199 { kX86 ## opname ## 32MC, kShiftMemCl, IS_LOAD | IS_STORE | IS_TERTIARY_OP | REG_USE0 | REG_USEC | SETS_CCODES, { 0, 0, 0xD3, 0, 0, modrm_opcode, 0, 0 }, #opname "32MC", "[!0r+!1d],cl" }, \
200 { kX86 ## opname ## 32AC, kShiftArrayCl, IS_LOAD | IS_STORE | IS_QUIN_OP | REG_USE01 | REG_USEC | SETS_CCODES, { 0, 0, 0xD3, 0, 0, modrm_opcode, 0, 0 }, #opname "32AC", "[!0r+!1r<<!2d+!3d],cl" }
201
202 SHIFT_ENCODING_MAP(Rol, 0x0),
203 SHIFT_ENCODING_MAP(Ror, 0x1),
204 SHIFT_ENCODING_MAP(Rcl, 0x2),
205 SHIFT_ENCODING_MAP(Rcr, 0x3),
206 SHIFT_ENCODING_MAP(Sal, 0x4),
207 SHIFT_ENCODING_MAP(Shr, 0x5),
208 SHIFT_ENCODING_MAP(Sar, 0x7),
209 #undef SHIFT_ENCODING_MAP
210
211 { kX86Cmc, kNullary, NO_OPERAND, { 0, 0, 0xF5, 0, 0, 0, 0, 0}, "Cmc", "" },
212
213 { kX86Test8RI, kRegImm, IS_BINARY_OP | REG_USE0 | SETS_CCODES, { 0, 0, 0xF6, 0, 0, 0, 0, 1}, "Test8RI", "!0r,!1d" },
214 { kX86Test8MI, kMemImm, IS_LOAD | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES, { 0, 0, 0xF6, 0, 0, 0, 0, 1}, "Test8MI", "[!0r+!1d],!2d" },
215 { kX86Test8AI, kArrayImm, IS_LOAD | IS_QUIN_OP | REG_USE01 | SETS_CCODES, { 0, 0, 0xF6, 0, 0, 0, 0, 1}, "Test8AI", "[!0r+!1r<<!2d+!3d],!4d" },
216 { kX86Test16RI, kRegImm, IS_BINARY_OP | REG_USE0 | SETS_CCODES, { 0x66, 0, 0xF7, 0, 0, 0, 0, 2}, "Test16RI", "!0r,!1d" },
217 { kX86Test16MI, kMemImm, IS_LOAD | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES, { 0x66, 0, 0xF7, 0, 0, 0, 0, 2}, "Test16MI", "[!0r+!1d],!2d" },
218 { kX86Test16AI, kArrayImm, IS_LOAD | IS_QUIN_OP | REG_USE01 | SETS_CCODES, { 0x66, 0, 0xF7, 0, 0, 0, 0, 2}, "Test16AI", "[!0r+!1r<<!2d+!3d],!4d" },
219 { kX86Test32RI, kRegImm, IS_BINARY_OP | REG_USE0 | SETS_CCODES, { 0, 0, 0xF7, 0, 0, 0, 0, 4}, "Test32RI", "!0r,!1d" },
220 { kX86Test32MI, kMemImm, IS_LOAD | IS_TERTIARY_OP | REG_USE0 | SETS_CCODES, { 0, 0, 0xF7, 0, 0, 0, 0, 4}, "Test32MI", "[!0r+!1d],!2d" },
221 { kX86Test32AI, kArrayImm, IS_LOAD | IS_QUIN_OP | REG_USE01 | SETS_CCODES, { 0, 0, 0xF7, 0, 0, 0, 0, 4}, "Test32AI", "[!0r+!1r<<!2d+!3d],!4d" },
222 { kX86Test32RR, kRegReg, IS_BINARY_OP | REG_USE01 | SETS_CCODES, { 0, 0, 0x85, 0, 0, 0, 0, 0}, "Test32RR", "!0r,!1r" },
223
224 #define UNARY_ENCODING_MAP(opname, modrm, is_store, sets_ccodes, \
225 reg, reg_kind, reg_flags, \
226 mem, mem_kind, mem_flags, \
227 arr, arr_kind, arr_flags, imm, \
228 b_flags, hw_flags, w_flags, \
229 b_format, hw_format, w_format) \
230 { kX86 ## opname ## 8 ## reg, reg_kind, reg_flags | b_flags | sets_ccodes, { 0, 0, 0xF6, 0, 0, modrm, 0, imm << 0}, #opname "8" #reg, #b_format "!0r" }, \
231 { kX86 ## opname ## 8 ## mem, mem_kind, IS_LOAD | is_store | mem_flags | b_flags | sets_ccodes, { 0, 0, 0xF6, 0, 0, modrm, 0, imm << 0}, #opname "8" #mem, #b_format "[!0r+!1d]" }, \
232 { kX86 ## opname ## 8 ## arr, arr_kind, IS_LOAD | is_store | arr_flags | b_flags | sets_ccodes, { 0, 0, 0xF6, 0, 0, modrm, 0, imm << 0}, #opname "8" #arr, #b_format "[!0r+!1r<<!2d+!3d]" }, \
233 { kX86 ## opname ## 16 ## reg, reg_kind, reg_flags | hw_flags | sets_ccodes, { 0x66, 0, 0xF7, 0, 0, modrm, 0, imm << 1}, #opname "16" #reg, #hw_format "!0r" }, \
234 { kX86 ## opname ## 16 ## mem, mem_kind, IS_LOAD | is_store | mem_flags | hw_flags | sets_ccodes, { 0x66, 0, 0xF7, 0, 0, modrm, 0, imm << 1}, #opname "16" #mem, #hw_format "[!0r+!1d]" }, \
235 { kX86 ## opname ## 16 ## arr, arr_kind, IS_LOAD | is_store | arr_flags | hw_flags | sets_ccodes, { 0x66, 0, 0xF7, 0, 0, modrm, 0, imm << 1}, #opname "16" #arr, #hw_format "[!0r+!1r<<!2d+!3d]" }, \
236 { kX86 ## opname ## 32 ## reg, reg_kind, reg_flags | w_flags | sets_ccodes, { 0, 0, 0xF7, 0, 0, modrm, 0, imm << 2}, #opname "32" #reg, #w_format "!0r" }, \
237 { kX86 ## opname ## 32 ## mem, mem_kind, IS_LOAD | is_store | mem_flags | w_flags | sets_ccodes, { 0, 0, 0xF7, 0, 0, modrm, 0, imm << 2}, #opname "32" #mem, #w_format "[!0r+!1d]" }, \
238 { kX86 ## opname ## 32 ## arr, arr_kind, IS_LOAD | is_store | arr_flags | w_flags | sets_ccodes, { 0, 0, 0xF7, 0, 0, modrm, 0, imm << 2}, #opname "32" #arr, #w_format "[!0r+!1r<<!2d+!3d]" }
239
240 UNARY_ENCODING_MAP(Not, 0x2, IS_STORE, 0, R, kReg, IS_UNARY_OP | REG_DEF0_USE0, M, kMem, IS_BINARY_OP | REG_USE0, A, kArray, IS_QUAD_OP | REG_USE01, 0, 0, 0, 0, "", "", ""),
241 UNARY_ENCODING_MAP(Neg, 0x3, IS_STORE, SETS_CCODES, R, kReg, IS_UNARY_OP | REG_DEF0_USE0, M, kMem, IS_BINARY_OP | REG_USE0, A, kArray, IS_QUAD_OP | REG_USE01, 0, 0, 0, 0, "", "", ""),
242
243 UNARY_ENCODING_MAP(Mul, 0x4, 0, SETS_CCODES, DaR, kRegRegReg, IS_UNARY_OP | REG_USE0, DaM, kRegRegMem, IS_BINARY_OP | REG_USE0, DaA, kRegRegArray, IS_QUAD_OP | REG_USE01, 0, REG_DEFA_USEA, REG_DEFAD_USEA, REG_DEFAD_USEA, "ax,al,", "dx:ax,ax,", "edx:eax,eax,"),
244 UNARY_ENCODING_MAP(Imul, 0x5, 0, SETS_CCODES, DaR, kRegRegReg, IS_UNARY_OP | REG_USE0, DaM, kRegRegMem, IS_BINARY_OP | REG_USE0, DaA, kRegRegArray, IS_QUAD_OP | REG_USE01, 0, REG_DEFA_USEA, REG_DEFAD_USEA, REG_DEFAD_USEA, "ax,al,", "dx:ax,ax,", "edx:eax,eax,"),
245 UNARY_ENCODING_MAP(Divmod, 0x6, 0, SETS_CCODES, DaR, kRegRegReg, IS_UNARY_OP | REG_USE0, DaM, kRegRegMem, IS_BINARY_OP | REG_USE0, DaA, kRegRegArray, IS_QUAD_OP | REG_USE01, 0, REG_DEFA_USEA, REG_DEFAD_USEAD, REG_DEFAD_USEAD, "ah:al,ax,", "dx:ax,dx:ax,", "edx:eax,edx:eax,"),
246 UNARY_ENCODING_MAP(Idivmod, 0x7, 0, SETS_CCODES, DaR, kRegRegReg, IS_UNARY_OP | REG_USE0, DaM, kRegRegMem, IS_BINARY_OP | REG_USE0, DaA, kRegRegArray, IS_QUAD_OP | REG_USE01, 0, REG_DEFA_USEA, REG_DEFAD_USEAD, REG_DEFAD_USEAD, "ah:al,ax,", "dx:ax,dx:ax,", "edx:eax,edx:eax,"),
247 #undef UNARY_ENCODING_MAP
248
249 #define EXT_0F_ENCODING_MAP(opname, prefix, opcode, reg_def) \
250 { kX86 ## opname ## RR, kRegReg, IS_BINARY_OP | reg_def | REG_USE01, { prefix, 0, 0x0F, opcode, 0, 0, 0, 0 }, #opname "RR", "!0r,!1r" }, \
251 { kX86 ## opname ## RM, kRegMem, IS_LOAD | IS_TERTIARY_OP | reg_def | REG_USE01, { prefix, 0, 0x0F, opcode, 0, 0, 0, 0 }, #opname "RM", "!0r,[!1r+!2d]" }, \
252 { kX86 ## opname ## RA, kRegArray, IS_LOAD | IS_QUIN_OP | reg_def | REG_USE012, { prefix, 0, 0x0F, opcode, 0, 0, 0, 0 }, #opname "RA", "!0r,[!1r+!2r<<!3d+!4d]" }
253
254 EXT_0F_ENCODING_MAP(Movsd, 0xF2, 0x10, REG_DEF0),
255 { kX86MovsdMR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02, { 0xF2, 0, 0x0F, 0x11, 0, 0, 0, 0 }, "MovsdMR", "[!0r+!1d],!2r" },
256 { kX86MovsdAR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014, { 0xF2, 0, 0x0F, 0x11, 0, 0, 0, 0 }, "MovsdAR", "[!0r+!1r<<!2d+!3d],!4r" },
257
258 EXT_0F_ENCODING_MAP(Movss, 0xF3, 0x10, REG_DEF0),
259 { kX86MovssMR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02, { 0xF3, 0, 0x0F, 0x11, 0, 0, 0, 0 }, "MovssMR", "[!0r+!1d],!2r" },
260 { kX86MovssAR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014, { 0xF3, 0, 0x0F, 0x11, 0, 0, 0, 0 }, "MovssAR", "[!0r+!1r<<!2d+!3d],!4r" },
261
262 EXT_0F_ENCODING_MAP(Cvtsi2sd, 0xF2, 0x2A, REG_DEF0),
263 EXT_0F_ENCODING_MAP(Cvtsi2ss, 0xF3, 0x2A, REG_DEF0),
264 EXT_0F_ENCODING_MAP(Cvttsd2si, 0xF2, 0x2C, REG_DEF0),
265 EXT_0F_ENCODING_MAP(Cvttss2si, 0xF3, 0x2C, REG_DEF0),
266 EXT_0F_ENCODING_MAP(Cvtsd2si, 0xF2, 0x2D, REG_DEF0),
267 EXT_0F_ENCODING_MAP(Cvtss2si, 0xF3, 0x2D, REG_DEF0),
268 EXT_0F_ENCODING_MAP(Ucomisd, 0x66, 0x2E, SETS_CCODES),
269 EXT_0F_ENCODING_MAP(Ucomiss, 0x00, 0x2E, SETS_CCODES),
270 EXT_0F_ENCODING_MAP(Comisd, 0x66, 0x2F, SETS_CCODES),
271 EXT_0F_ENCODING_MAP(Comiss, 0x00, 0x2F, SETS_CCODES),
272 EXT_0F_ENCODING_MAP(Orps, 0x00, 0x56, REG_DEF0),
273 EXT_0F_ENCODING_MAP(Xorps, 0x00, 0x57, REG_DEF0),
274 EXT_0F_ENCODING_MAP(Addsd, 0xF2, 0x58, REG_DEF0),
275 EXT_0F_ENCODING_MAP(Addss, 0xF3, 0x58, REG_DEF0),
276 EXT_0F_ENCODING_MAP(Mulsd, 0xF2, 0x59, REG_DEF0),
277 EXT_0F_ENCODING_MAP(Mulss, 0xF3, 0x59, REG_DEF0),
278 EXT_0F_ENCODING_MAP(Cvtsd2ss, 0xF2, 0x5A, REG_DEF0),
279 EXT_0F_ENCODING_MAP(Cvtss2sd, 0xF3, 0x5A, REG_DEF0),
280 EXT_0F_ENCODING_MAP(Subsd, 0xF2, 0x5C, REG_DEF0),
281 EXT_0F_ENCODING_MAP(Subss, 0xF3, 0x5C, REG_DEF0),
282 EXT_0F_ENCODING_MAP(Divsd, 0xF2, 0x5E, REG_DEF0),
283 EXT_0F_ENCODING_MAP(Divss, 0xF3, 0x5E, REG_DEF0),
284
285 { kX86PsrlqRI, kRegImm, IS_BINARY_OP | REG_DEF0_USE0, { 0x66, 0, 0x0F, 0x73, 0, 2, 0, 1 }, "PsrlqRI", "!0r,!1d" },
286 { kX86PsllqRI, kRegImm, IS_BINARY_OP | REG_DEF0_USE0, { 0x66, 0, 0x0F, 0x73, 0, 6, 0, 1 }, "PsllqRI", "!0r,!1d" },
287
288 EXT_0F_ENCODING_MAP(Movdxr, 0x66, 0x6E, REG_DEF0),
289 { kX86MovdrxRR, kRegRegStore, IS_BINARY_OP | REG_DEF0 | REG_USE01, { 0x66, 0, 0x0F, 0x7E, 0, 0, 0, 0 }, "MovdrxRR", "!0r,!1r" },
290 { kX86MovdrxMR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02, { 0x66, 0, 0x0F, 0x7E, 0, 0, 0, 0 }, "MovdrxMR", "[!0r+!1d],!2r" },
291 { kX86MovdrxAR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014, { 0x66, 0, 0x0F, 0x7E, 0, 0, 0, 0 }, "MovdrxAR", "[!0r+!1r<<!2d+!3d],!4r" },
292
293 { kX86Set8R, kRegCond, IS_BINARY_OP | REG_DEF0 | USES_CCODES, { 0, 0, 0x0F, 0x90, 0, 0, 0, 0 }, "Set8R", "!1c !0r" },
294 { kX86Set8M, kMemCond, IS_STORE | IS_TERTIARY_OP | REG_USE0 | USES_CCODES, { 0, 0, 0x0F, 0x90, 0, 0, 0, 0 }, "Set8M", "!2c [!0r+!1d]" },
295 { kX86Set8A, kArrayCond, IS_STORE | IS_QUIN_OP | REG_USE01 | USES_CCODES, { 0, 0, 0x0F, 0x90, 0, 0, 0, 0 }, "Set8A", "!4c [!0r+!1r<<!2d+!3d]" },
296
297 // TODO: load/store?
298 // Encode the modrm opcode as an extra opcode byte to avoid computation during assembly.
299 { kX86Mfence, kReg, NO_OPERAND, { 0, 0, 0x0F, 0xAE, 0, 6, 0, 0 }, "Mfence", "" },
300
301 EXT_0F_ENCODING_MAP(Imul16, 0x66, 0xAF, REG_DEF0 | SETS_CCODES),
302 EXT_0F_ENCODING_MAP(Imul32, 0x00, 0xAF, REG_DEF0 | SETS_CCODES),
303
304 { kX86CmpxchgRR, kRegRegStore, IS_BINARY_OP | REG_DEF0 | REG_USE01 | REG_DEFA_USEA | SETS_CCODES, { 0, 0, 0x0F, 0xB1, 0, 0, 0, 0 }, "Cmpxchg", "!0r,!1r" },
305 { kX86CmpxchgMR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02 | REG_DEFA_USEA | SETS_CCODES, { 0, 0, 0x0F, 0xB1, 0, 0, 0, 0 }, "Cmpxchg", "[!0r+!1d],!2r" },
306 { kX86CmpxchgAR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014 | REG_DEFA_USEA | SETS_CCODES, { 0, 0, 0x0F, 0xB1, 0, 0, 0, 0 }, "Cmpxchg", "[!0r+!1r<<!2d+!3d],!4r" },
307 { kX86LockCmpxchgRR, kRegRegStore, IS_BINARY_OP | REG_DEF0 | REG_USE01 | REG_DEFA_USEA | SETS_CCODES, { 0xF0, 0, 0x0F, 0xB1, 0, 0, 0, 0 }, "Lock Cmpxchg", "!0r,!1r" },
308 { kX86LockCmpxchgMR, kMemReg, IS_STORE | IS_TERTIARY_OP | REG_USE02 | REG_DEFA_USEA | SETS_CCODES, { 0xF0, 0, 0x0F, 0xB1, 0, 0, 0, 0 }, "Lock Cmpxchg", "[!0r+!1d],!2r" },
309 { kX86LockCmpxchgAR, kArrayReg, IS_STORE | IS_QUIN_OP | REG_USE014 | REG_DEFA_USEA | SETS_CCODES, { 0xF0, 0, 0x0F, 0xB1, 0, 0, 0, 0 }, "Lock Cmpxchg", "[!0r+!1r<<!2d+!3d],!4r" },
310
311 EXT_0F_ENCODING_MAP(Movzx8, 0x00, 0xB6, REG_DEF0),
312 EXT_0F_ENCODING_MAP(Movzx16, 0x00, 0xB7, REG_DEF0),
313 EXT_0F_ENCODING_MAP(Movsx8, 0x00, 0xBE, REG_DEF0),
314 EXT_0F_ENCODING_MAP(Movsx16, 0x00, 0xBF, REG_DEF0),
315 #undef EXT_0F_ENCODING_MAP
316
317 { kX86Jcc8, kJcc, IS_BINARY_OP | IS_BRANCH | NEEDS_FIXUP | USES_CCODES, { 0, 0, 0x70, 0, 0, 0, 0, 0 }, "Jcc8", "!1c !0t" },
318 { kX86Jcc32, kJcc, IS_BINARY_OP | IS_BRANCH | NEEDS_FIXUP | USES_CCODES, { 0, 0, 0x0F, 0x80, 0, 0, 0, 0 }, "Jcc32", "!1c !0t" },
319 { kX86Jmp8, kJmp, IS_UNARY_OP | IS_BRANCH | NEEDS_FIXUP, { 0, 0, 0xEB, 0, 0, 0, 0, 0 }, "Jmp8", "!0t" },
320 { kX86Jmp32, kJmp, IS_UNARY_OP | IS_BRANCH | NEEDS_FIXUP, { 0, 0, 0xE9, 0, 0, 0, 0, 0 }, "Jmp32", "!0t" },
321 { kX86JmpR, kJmp, IS_UNARY_OP | IS_BRANCH | REG_USE0, { 0, 0, 0xFF, 0, 0, 4, 0, 0 }, "JmpR", "!0r" },
322 { kX86JmpT, kJmp, IS_UNARY_OP | IS_BRANCH | IS_LOAD, { THREAD_PREFIX, 0, 0xFF, 0, 0, 4, 0, 0 }, "JmpT", "fs:[!0d]" },
323 { kX86CallR, kCall, IS_UNARY_OP | IS_BRANCH | REG_USE0, { 0, 0, 0xE8, 0, 0, 0, 0, 0 }, "CallR", "!0r" },
324 { kX86CallM, kCall, IS_BINARY_OP | IS_BRANCH | IS_LOAD | REG_USE0, { 0, 0, 0xFF, 0, 0, 2, 0, 0 }, "CallM", "[!0r+!1d]" },
325 { kX86CallA, kCall, IS_QUAD_OP | IS_BRANCH | IS_LOAD | REG_USE01, { 0, 0, 0xFF, 0, 0, 2, 0, 0 }, "CallA", "[!0r+!1r<<!2d+!3d]" },
326 { kX86CallT, kCall, IS_UNARY_OP | IS_BRANCH | IS_LOAD, { THREAD_PREFIX, 0, 0xFF, 0, 0, 2, 0, 0 }, "CallT", "fs:[!0d]" },
327 { kX86Ret, kNullary, NO_OPERAND | IS_BRANCH, { 0, 0, 0xC3, 0, 0, 0, 0, 0 }, "Ret", "" },
328
329 { kX86StartOfMethod, kMacro, IS_UNARY_OP | SETS_CCODES, { 0, 0, 0, 0, 0, 0, 0, 0 }, "StartOfMethod", "!0r" },
330 { kX86PcRelLoadRA, kPcRel, IS_LOAD | IS_QUIN_OP | REG_DEF0_USE12, { 0, 0, 0x8B, 0, 0, 0, 0, 0 }, "PcRelLoadRA", "!0r,[!1r+!2r<<!3d+!4p]" },
331 { kX86PcRelAdr, kPcRel, IS_LOAD | IS_BINARY_OP | REG_DEF0, { 0, 0, 0xB8, 0, 0, 0, 0, 4 }, "PcRelAdr", "!0r,!1d" },
332 };
333
ComputeSize(const X86EncodingMap * entry,int base,int displacement,bool has_sib)334 static size_t ComputeSize(const X86EncodingMap* entry, int base, int displacement, bool has_sib) {
335 size_t size = 0;
336 if (entry->skeleton.prefix1 > 0) {
337 ++size;
338 if (entry->skeleton.prefix2 > 0) {
339 ++size;
340 }
341 }
342 ++size; // opcode
343 if (entry->skeleton.opcode == 0x0F) {
344 ++size;
345 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode1 == 0x3A) {
346 ++size;
347 }
348 }
349 ++size; // modrm
350 if (has_sib || base == rX86_SP) {
351 // SP requires a SIB byte.
352 ++size;
353 }
354 if (displacement != 0 || base == rBP) {
355 // BP requires an explicit displacement, even when it's 0.
356 if (entry->opcode != kX86Lea32RA) {
357 DCHECK_NE(entry->flags & (IS_LOAD | IS_STORE), 0ULL) << entry->name;
358 }
359 size += IS_SIMM8(displacement) ? 1 : 4;
360 }
361 size += entry->skeleton.immediate_bytes;
362 return size;
363 }
364
GetInsnSize(LIR * lir)365 int X86Mir2Lir::GetInsnSize(LIR* lir) {
366 const X86EncodingMap* entry = &X86Mir2Lir::EncodingMap[lir->opcode];
367 switch (entry->kind) {
368 case kData:
369 return 4; // 4 bytes of data
370 case kNop:
371 return lir->operands[0]; // length of nop is sole operand
372 case kNullary:
373 return 1; // 1 byte of opcode
374 case kReg: // lir operands - 0: reg
375 return ComputeSize(entry, 0, 0, false);
376 case kMem: // lir operands - 0: base, 1: disp
377 return ComputeSize(entry, lir->operands[0], lir->operands[1], false);
378 case kArray: // lir operands - 0: base, 1: index, 2: scale, 3: disp
379 return ComputeSize(entry, lir->operands[0], lir->operands[3], true);
380 case kMemReg: // lir operands - 0: base, 1: disp, 2: reg
381 return ComputeSize(entry, lir->operands[0], lir->operands[1], false);
382 case kArrayReg: // lir operands - 0: base, 1: index, 2: scale, 3: disp, 4: reg
383 return ComputeSize(entry, lir->operands[0], lir->operands[3], true);
384 case kThreadReg: // lir operands - 0: disp, 1: reg
385 return ComputeSize(entry, 0, lir->operands[0], false);
386 case kRegReg:
387 return ComputeSize(entry, 0, 0, false);
388 case kRegRegStore:
389 return ComputeSize(entry, 0, 0, false);
390 case kRegMem: // lir operands - 0: reg, 1: base, 2: disp
391 return ComputeSize(entry, lir->operands[1], lir->operands[2], false);
392 case kRegArray: // lir operands - 0: reg, 1: base, 2: index, 3: scale, 4: disp
393 return ComputeSize(entry, lir->operands[1], lir->operands[4], true);
394 case kRegThread: // lir operands - 0: reg, 1: disp
395 return ComputeSize(entry, 0, 0x12345678, false); // displacement size is always 32bit
396 case kRegImm: { // lir operands - 0: reg, 1: immediate
397 size_t size = ComputeSize(entry, 0, 0, false);
398 if (entry->skeleton.ax_opcode == 0) {
399 return size;
400 } else {
401 // AX opcodes don't require the modrm byte.
402 int reg = lir->operands[0];
403 return size - (reg == rAX ? 1 : 0);
404 }
405 }
406 case kMemImm: // lir operands - 0: base, 1: disp, 2: immediate
407 return ComputeSize(entry, lir->operands[0], lir->operands[1], false);
408 case kArrayImm: // lir operands - 0: base, 1: index, 2: scale, 3: disp 4: immediate
409 return ComputeSize(entry, lir->operands[0], lir->operands[3], true);
410 case kThreadImm: // lir operands - 0: disp, 1: imm
411 return ComputeSize(entry, 0, 0x12345678, false); // displacement size is always 32bit
412 case kRegRegImm: // lir operands - 0: reg, 1: reg, 2: imm
413 return ComputeSize(entry, 0, 0, false);
414 case kRegMemImm: // lir operands - 0: reg, 1: base, 2: disp, 3: imm
415 return ComputeSize(entry, lir->operands[1], lir->operands[2], false);
416 case kRegArrayImm: // lir operands - 0: reg, 1: base, 2: index, 3: scale, 4: disp, 5: imm
417 return ComputeSize(entry, lir->operands[1], lir->operands[4], true);
418 case kMovRegImm: // lir operands - 0: reg, 1: immediate
419 return 1 + entry->skeleton.immediate_bytes;
420 case kShiftRegImm: // lir operands - 0: reg, 1: immediate
421 // Shift by immediate one has a shorter opcode.
422 return ComputeSize(entry, 0, 0, false) - (lir->operands[1] == 1 ? 1 : 0);
423 case kShiftMemImm: // lir operands - 0: base, 1: disp, 2: immediate
424 // Shift by immediate one has a shorter opcode.
425 return ComputeSize(entry, lir->operands[0], lir->operands[1], false) -
426 (lir->operands[2] == 1 ? 1 : 0);
427 case kShiftArrayImm: // lir operands - 0: base, 1: index, 2: scale, 3: disp 4: immediate
428 // Shift by immediate one has a shorter opcode.
429 return ComputeSize(entry, lir->operands[0], lir->operands[3], true) -
430 (lir->operands[4] == 1 ? 1 : 0);
431 case kShiftRegCl:
432 return ComputeSize(entry, 0, 0, false);
433 case kShiftMemCl: // lir operands - 0: base, 1: disp, 2: cl
434 return ComputeSize(entry, lir->operands[0], lir->operands[1], false);
435 case kShiftArrayCl: // lir operands - 0: base, 1: index, 2: scale, 3: disp, 4: reg
436 return ComputeSize(entry, lir->operands[0], lir->operands[3], true);
437 case kRegCond: // lir operands - 0: reg, 1: cond
438 return ComputeSize(entry, 0, 0, false);
439 case kMemCond: // lir operands - 0: base, 1: disp, 2: cond
440 return ComputeSize(entry, lir->operands[0], lir->operands[1], false);
441 case kArrayCond: // lir operands - 0: base, 1: index, 2: scale, 3: disp, 4: cond
442 return ComputeSize(entry, lir->operands[0], lir->operands[3], true);
443 case kJcc:
444 if (lir->opcode == kX86Jcc8) {
445 return 2; // opcode + rel8
446 } else {
447 DCHECK(lir->opcode == kX86Jcc32);
448 return 6; // 2 byte opcode + rel32
449 }
450 case kJmp:
451 if (lir->opcode == kX86Jmp8) {
452 return 2; // opcode + rel8
453 } else if (lir->opcode == kX86Jmp32) {
454 return 5; // opcode + rel32
455 } else if (lir->opcode == kX86JmpT) {
456 return ComputeSize(entry, 0, 0x12345678, false); // displacement size is always 32bit
457 } else {
458 DCHECK(lir->opcode == kX86JmpR);
459 return 2; // opcode + modrm
460 }
461 case kCall:
462 switch (lir->opcode) {
463 case kX86CallR: return 2; // opcode modrm
464 case kX86CallM: // lir operands - 0: base, 1: disp
465 return ComputeSize(entry, lir->operands[0], lir->operands[1], false);
466 case kX86CallA: // lir operands - 0: base, 1: index, 2: scale, 3: disp
467 return ComputeSize(entry, lir->operands[0], lir->operands[3], true);
468 case kX86CallT: // lir operands - 0: disp
469 return ComputeSize(entry, 0, 0x12345678, false); // displacement size is always 32bit
470 default:
471 break;
472 }
473 break;
474 case kPcRel:
475 if (entry->opcode == kX86PcRelLoadRA) {
476 // lir operands - 0: reg, 1: base, 2: index, 3: scale, 4: table
477 return ComputeSize(entry, lir->operands[1], 0x12345678, true);
478 } else {
479 DCHECK(entry->opcode == kX86PcRelAdr);
480 return 5; // opcode with reg + 4 byte immediate
481 }
482 case kMacro:
483 DCHECK_EQ(lir->opcode, static_cast<int>(kX86StartOfMethod));
484 return 5 /* call opcode + 4 byte displacement */ + 1 /* pop reg */ +
485 ComputeSize(&X86Mir2Lir::EncodingMap[kX86Sub32RI], 0, 0, false) -
486 (lir->operands[0] == rAX ? 1 : 0); // shorter ax encoding
487 default:
488 break;
489 }
490 UNIMPLEMENTED(FATAL) << "Unimplemented size encoding for: " << entry->name;
491 return 0;
492 }
493
ModrmForDisp(int base,int disp)494 static uint8_t ModrmForDisp(int base, int disp) {
495 // BP requires an explicit disp, so do not omit it in the 0 case
496 if (disp == 0 && base != rBP) {
497 return 0;
498 } else if (IS_SIMM8(disp)) {
499 return 1;
500 } else {
501 return 2;
502 }
503 }
504
EmitDisp(int base,int disp)505 void X86Mir2Lir::EmitDisp(int base, int disp) {
506 // BP requires an explicit disp, so do not omit it in the 0 case
507 if (disp == 0 && base != rBP) {
508 return;
509 } else if (IS_SIMM8(disp)) {
510 code_buffer_.push_back(disp & 0xFF);
511 } else {
512 code_buffer_.push_back(disp & 0xFF);
513 code_buffer_.push_back((disp >> 8) & 0xFF);
514 code_buffer_.push_back((disp >> 16) & 0xFF);
515 code_buffer_.push_back((disp >> 24) & 0xFF);
516 }
517 }
518
EmitOpReg(const X86EncodingMap * entry,uint8_t reg)519 void X86Mir2Lir::EmitOpReg(const X86EncodingMap* entry, uint8_t reg) {
520 if (entry->skeleton.prefix1 != 0) {
521 code_buffer_.push_back(entry->skeleton.prefix1);
522 if (entry->skeleton.prefix2 != 0) {
523 code_buffer_.push_back(entry->skeleton.prefix2);
524 }
525 } else {
526 DCHECK_EQ(0, entry->skeleton.prefix2);
527 }
528 code_buffer_.push_back(entry->skeleton.opcode);
529 if (entry->skeleton.opcode == 0x0F) {
530 code_buffer_.push_back(entry->skeleton.extra_opcode1);
531 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
532 code_buffer_.push_back(entry->skeleton.extra_opcode2);
533 } else {
534 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
535 }
536 } else {
537 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
538 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
539 }
540 if (X86_FPREG(reg)) {
541 reg = reg & X86_FP_REG_MASK;
542 }
543 if (reg >= 4) {
544 DCHECK(strchr(entry->name, '8') == NULL) << entry->name << " " << static_cast<int>(reg)
545 << " in " << PrettyMethod(cu_->method_idx, *cu_->dex_file);
546 }
547 DCHECK_LT(reg, 8);
548 uint8_t modrm = (3 << 6) | (entry->skeleton.modrm_opcode << 3) | reg;
549 code_buffer_.push_back(modrm);
550 DCHECK_EQ(0, entry->skeleton.ax_opcode);
551 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
552 }
553
EmitOpMem(const X86EncodingMap * entry,uint8_t base,int disp)554 void X86Mir2Lir::EmitOpMem(const X86EncodingMap* entry, uint8_t base, int disp) {
555 if (entry->skeleton.prefix1 != 0) {
556 code_buffer_.push_back(entry->skeleton.prefix1);
557 if (entry->skeleton.prefix2 != 0) {
558 code_buffer_.push_back(entry->skeleton.prefix2);
559 }
560 } else {
561 DCHECK_EQ(0, entry->skeleton.prefix2);
562 }
563 code_buffer_.push_back(entry->skeleton.opcode);
564 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
565 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
566 DCHECK_LT(entry->skeleton.modrm_opcode, 8);
567 DCHECK_LT(base, 8);
568 uint8_t modrm = (ModrmForDisp(base, disp) << 6) | (entry->skeleton.modrm_opcode << 3) | base;
569 code_buffer_.push_back(modrm);
570 EmitDisp(base, disp);
571 DCHECK_EQ(0, entry->skeleton.ax_opcode);
572 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
573 }
574
EmitMemReg(const X86EncodingMap * entry,uint8_t base,int disp,uint8_t reg)575 void X86Mir2Lir::EmitMemReg(const X86EncodingMap* entry,
576 uint8_t base, int disp, uint8_t reg) {
577 if (entry->skeleton.prefix1 != 0) {
578 code_buffer_.push_back(entry->skeleton.prefix1);
579 if (entry->skeleton.prefix2 != 0) {
580 code_buffer_.push_back(entry->skeleton.prefix2);
581 }
582 } else {
583 DCHECK_EQ(0, entry->skeleton.prefix2);
584 }
585 code_buffer_.push_back(entry->skeleton.opcode);
586 if (entry->skeleton.opcode == 0x0F) {
587 code_buffer_.push_back(entry->skeleton.extra_opcode1);
588 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
589 code_buffer_.push_back(entry->skeleton.extra_opcode2);
590 } else {
591 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
592 }
593 } else {
594 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
595 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
596 }
597 if (X86_FPREG(reg)) {
598 reg = reg & X86_FP_REG_MASK;
599 }
600 if (reg >= 4) {
601 DCHECK(strchr(entry->name, '8') == NULL) << entry->name << " " << static_cast<int>(reg)
602 << " in " << PrettyMethod(cu_->method_idx, *cu_->dex_file);
603 }
604 DCHECK_LT(reg, 8);
605 DCHECK_LT(base, 8);
606 uint8_t modrm = (ModrmForDisp(base, disp) << 6) | (reg << 3) | base;
607 code_buffer_.push_back(modrm);
608 if (base == rX86_SP) {
609 // Special SIB for SP base
610 code_buffer_.push_back(0 << 6 | (rX86_SP << 3) | rX86_SP);
611 }
612 EmitDisp(base, disp);
613 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
614 DCHECK_EQ(0, entry->skeleton.ax_opcode);
615 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
616 }
617
EmitRegMem(const X86EncodingMap * entry,uint8_t reg,uint8_t base,int disp)618 void X86Mir2Lir::EmitRegMem(const X86EncodingMap* entry,
619 uint8_t reg, uint8_t base, int disp) {
620 // Opcode will flip operands.
621 EmitMemReg(entry, base, disp, reg);
622 }
623
EmitRegArray(const X86EncodingMap * entry,uint8_t reg,uint8_t base,uint8_t index,int scale,int disp)624 void X86Mir2Lir::EmitRegArray(const X86EncodingMap* entry, uint8_t reg, uint8_t base, uint8_t index,
625 int scale, int disp) {
626 if (entry->skeleton.prefix1 != 0) {
627 code_buffer_.push_back(entry->skeleton.prefix1);
628 if (entry->skeleton.prefix2 != 0) {
629 code_buffer_.push_back(entry->skeleton.prefix2);
630 }
631 } else {
632 DCHECK_EQ(0, entry->skeleton.prefix2);
633 }
634 code_buffer_.push_back(entry->skeleton.opcode);
635 if (entry->skeleton.opcode == 0x0F) {
636 code_buffer_.push_back(entry->skeleton.extra_opcode1);
637 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
638 code_buffer_.push_back(entry->skeleton.extra_opcode2);
639 } else {
640 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
641 }
642 } else {
643 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
644 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
645 }
646 if (X86_FPREG(reg)) {
647 reg = reg & X86_FP_REG_MASK;
648 }
649 DCHECK_LT(reg, 8);
650 uint8_t modrm = (ModrmForDisp(base, disp) << 6) | (reg << 3) | rX86_SP;
651 code_buffer_.push_back(modrm);
652 DCHECK_LT(scale, 4);
653 DCHECK_LT(index, 8);
654 DCHECK_LT(base, 8);
655 uint8_t sib = (scale << 6) | (index << 3) | base;
656 code_buffer_.push_back(sib);
657 EmitDisp(base, disp);
658 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
659 DCHECK_EQ(0, entry->skeleton.ax_opcode);
660 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
661 }
662
EmitArrayReg(const X86EncodingMap * entry,uint8_t base,uint8_t index,int scale,int disp,uint8_t reg)663 void X86Mir2Lir::EmitArrayReg(const X86EncodingMap* entry, uint8_t base, uint8_t index, int scale, int disp,
664 uint8_t reg) {
665 // Opcode will flip operands.
666 EmitRegArray(entry, reg, base, index, scale, disp);
667 }
668
EmitRegThread(const X86EncodingMap * entry,uint8_t reg,int disp)669 void X86Mir2Lir::EmitRegThread(const X86EncodingMap* entry, uint8_t reg, int disp) {
670 DCHECK_NE(entry->skeleton.prefix1, 0);
671 code_buffer_.push_back(entry->skeleton.prefix1);
672 if (entry->skeleton.prefix2 != 0) {
673 code_buffer_.push_back(entry->skeleton.prefix2);
674 }
675 code_buffer_.push_back(entry->skeleton.opcode);
676 if (entry->skeleton.opcode == 0x0F) {
677 code_buffer_.push_back(entry->skeleton.extra_opcode1);
678 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
679 code_buffer_.push_back(entry->skeleton.extra_opcode2);
680 } else {
681 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
682 }
683 } else {
684 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
685 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
686 }
687 if (X86_FPREG(reg)) {
688 reg = reg & X86_FP_REG_MASK;
689 }
690 if (reg >= 4) {
691 DCHECK(strchr(entry->name, '8') == NULL) << entry->name << " " << static_cast<int>(reg)
692 << " in " << PrettyMethod(cu_->method_idx, *cu_->dex_file);
693 }
694 DCHECK_LT(reg, 8);
695 uint8_t modrm = (0 << 6) | (reg << 3) | rBP;
696 code_buffer_.push_back(modrm);
697 code_buffer_.push_back(disp & 0xFF);
698 code_buffer_.push_back((disp >> 8) & 0xFF);
699 code_buffer_.push_back((disp >> 16) & 0xFF);
700 code_buffer_.push_back((disp >> 24) & 0xFF);
701 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
702 DCHECK_EQ(0, entry->skeleton.ax_opcode);
703 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
704 }
705
EmitRegReg(const X86EncodingMap * entry,uint8_t reg1,uint8_t reg2)706 void X86Mir2Lir::EmitRegReg(const X86EncodingMap* entry, uint8_t reg1, uint8_t reg2) {
707 if (entry->skeleton.prefix1 != 0) {
708 code_buffer_.push_back(entry->skeleton.prefix1);
709 if (entry->skeleton.prefix2 != 0) {
710 code_buffer_.push_back(entry->skeleton.prefix2);
711 }
712 } else {
713 DCHECK_EQ(0, entry->skeleton.prefix2);
714 }
715 code_buffer_.push_back(entry->skeleton.opcode);
716 if (entry->skeleton.opcode == 0x0F) {
717 code_buffer_.push_back(entry->skeleton.extra_opcode1);
718 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
719 code_buffer_.push_back(entry->skeleton.extra_opcode2);
720 } else {
721 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
722 }
723 } else {
724 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
725 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
726 }
727 if (X86_FPREG(reg1)) {
728 reg1 = reg1 & X86_FP_REG_MASK;
729 }
730 if (X86_FPREG(reg2)) {
731 reg2 = reg2 & X86_FP_REG_MASK;
732 }
733 DCHECK_LT(reg1, 8);
734 DCHECK_LT(reg2, 8);
735 uint8_t modrm = (3 << 6) | (reg1 << 3) | reg2;
736 code_buffer_.push_back(modrm);
737 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
738 DCHECK_EQ(0, entry->skeleton.ax_opcode);
739 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
740 }
741
EmitRegRegImm(const X86EncodingMap * entry,uint8_t reg1,uint8_t reg2,int32_t imm)742 void X86Mir2Lir::EmitRegRegImm(const X86EncodingMap* entry,
743 uint8_t reg1, uint8_t reg2, int32_t imm) {
744 if (entry->skeleton.prefix1 != 0) {
745 code_buffer_.push_back(entry->skeleton.prefix1);
746 if (entry->skeleton.prefix2 != 0) {
747 code_buffer_.push_back(entry->skeleton.prefix2);
748 }
749 } else {
750 DCHECK_EQ(0, entry->skeleton.prefix2);
751 }
752 code_buffer_.push_back(entry->skeleton.opcode);
753 if (entry->skeleton.opcode == 0x0F) {
754 code_buffer_.push_back(entry->skeleton.extra_opcode1);
755 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
756 code_buffer_.push_back(entry->skeleton.extra_opcode2);
757 } else {
758 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
759 }
760 } else {
761 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
762 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
763 }
764 if (X86_FPREG(reg1)) {
765 reg1 = reg1 & X86_FP_REG_MASK;
766 }
767 if (X86_FPREG(reg2)) {
768 reg2 = reg2 & X86_FP_REG_MASK;
769 }
770 DCHECK_LT(reg1, 8);
771 DCHECK_LT(reg2, 8);
772 uint8_t modrm = (3 << 6) | (reg1 << 3) | reg2;
773 code_buffer_.push_back(modrm);
774 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
775 DCHECK_EQ(0, entry->skeleton.ax_opcode);
776 switch (entry->skeleton.immediate_bytes) {
777 case 1:
778 DCHECK(IS_SIMM8(imm));
779 code_buffer_.push_back(imm & 0xFF);
780 break;
781 case 2:
782 DCHECK(IS_SIMM16(imm));
783 code_buffer_.push_back(imm & 0xFF);
784 code_buffer_.push_back((imm >> 8) & 0xFF);
785 break;
786 case 4:
787 code_buffer_.push_back(imm & 0xFF);
788 code_buffer_.push_back((imm >> 8) & 0xFF);
789 code_buffer_.push_back((imm >> 16) & 0xFF);
790 code_buffer_.push_back((imm >> 24) & 0xFF);
791 break;
792 default:
793 LOG(FATAL) << "Unexpected immediate bytes (" << entry->skeleton.immediate_bytes
794 << ") for instruction: " << entry->name;
795 break;
796 }
797 }
798
EmitRegImm(const X86EncodingMap * entry,uint8_t reg,int imm)799 void X86Mir2Lir::EmitRegImm(const X86EncodingMap* entry, uint8_t reg, int imm) {
800 if (entry->skeleton.prefix1 != 0) {
801 code_buffer_.push_back(entry->skeleton.prefix1);
802 if (entry->skeleton.prefix2 != 0) {
803 code_buffer_.push_back(entry->skeleton.prefix2);
804 }
805 } else {
806 DCHECK_EQ(0, entry->skeleton.prefix2);
807 }
808 if (reg == rAX && entry->skeleton.ax_opcode != 0) {
809 code_buffer_.push_back(entry->skeleton.ax_opcode);
810 } else {
811 code_buffer_.push_back(entry->skeleton.opcode);
812 if (entry->skeleton.opcode == 0x0F) {
813 code_buffer_.push_back(entry->skeleton.extra_opcode1);
814 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
815 code_buffer_.push_back(entry->skeleton.extra_opcode2);
816 } else {
817 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
818 }
819 } else {
820 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
821 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
822 }
823 if (X86_FPREG(reg)) {
824 reg = reg & X86_FP_REG_MASK;
825 }
826 uint8_t modrm = (3 << 6) | (entry->skeleton.modrm_opcode << 3) | reg;
827 code_buffer_.push_back(modrm);
828 }
829 switch (entry->skeleton.immediate_bytes) {
830 case 1:
831 DCHECK(IS_SIMM8(imm));
832 code_buffer_.push_back(imm & 0xFF);
833 break;
834 case 2:
835 DCHECK(IS_SIMM16(imm));
836 code_buffer_.push_back(imm & 0xFF);
837 code_buffer_.push_back((imm >> 8) & 0xFF);
838 break;
839 case 4:
840 code_buffer_.push_back(imm & 0xFF);
841 code_buffer_.push_back((imm >> 8) & 0xFF);
842 code_buffer_.push_back((imm >> 16) & 0xFF);
843 code_buffer_.push_back((imm >> 24) & 0xFF);
844 break;
845 default:
846 LOG(FATAL) << "Unexpected immediate bytes (" << entry->skeleton.immediate_bytes
847 << ") for instruction: " << entry->name;
848 break;
849 }
850 }
851
EmitThreadImm(const X86EncodingMap * entry,int disp,int imm)852 void X86Mir2Lir::EmitThreadImm(const X86EncodingMap* entry, int disp, int imm) {
853 if (entry->skeleton.prefix1 != 0) {
854 code_buffer_.push_back(entry->skeleton.prefix1);
855 if (entry->skeleton.prefix2 != 0) {
856 code_buffer_.push_back(entry->skeleton.prefix2);
857 }
858 } else {
859 DCHECK_EQ(0, entry->skeleton.prefix2);
860 }
861 code_buffer_.push_back(entry->skeleton.opcode);
862 if (entry->skeleton.opcode == 0x0F) {
863 code_buffer_.push_back(entry->skeleton.extra_opcode1);
864 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
865 code_buffer_.push_back(entry->skeleton.extra_opcode2);
866 } else {
867 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
868 }
869 } else {
870 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
871 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
872 }
873 uint8_t modrm = (0 << 6) | (entry->skeleton.modrm_opcode << 3) | rBP;
874 code_buffer_.push_back(modrm);
875 code_buffer_.push_back(disp & 0xFF);
876 code_buffer_.push_back((disp >> 8) & 0xFF);
877 code_buffer_.push_back((disp >> 16) & 0xFF);
878 code_buffer_.push_back((disp >> 24) & 0xFF);
879 switch (entry->skeleton.immediate_bytes) {
880 case 1:
881 DCHECK(IS_SIMM8(imm));
882 code_buffer_.push_back(imm & 0xFF);
883 break;
884 case 2:
885 DCHECK(IS_SIMM16(imm));
886 code_buffer_.push_back(imm & 0xFF);
887 code_buffer_.push_back((imm >> 8) & 0xFF);
888 break;
889 case 4:
890 code_buffer_.push_back(imm & 0xFF);
891 code_buffer_.push_back((imm >> 8) & 0xFF);
892 code_buffer_.push_back((imm >> 16) & 0xFF);
893 code_buffer_.push_back((imm >> 24) & 0xFF);
894 break;
895 default:
896 LOG(FATAL) << "Unexpected immediate bytes (" << entry->skeleton.immediate_bytes
897 << ") for instruction: " << entry->name;
898 break;
899 }
900 DCHECK_EQ(entry->skeleton.ax_opcode, 0);
901 }
902
EmitMovRegImm(const X86EncodingMap * entry,uint8_t reg,int imm)903 void X86Mir2Lir::EmitMovRegImm(const X86EncodingMap* entry, uint8_t reg, int imm) {
904 DCHECK_LT(reg, 8);
905 code_buffer_.push_back(0xB8 + reg);
906 code_buffer_.push_back(imm & 0xFF);
907 code_buffer_.push_back((imm >> 8) & 0xFF);
908 code_buffer_.push_back((imm >> 16) & 0xFF);
909 code_buffer_.push_back((imm >> 24) & 0xFF);
910 }
911
EmitShiftRegImm(const X86EncodingMap * entry,uint8_t reg,int imm)912 void X86Mir2Lir::EmitShiftRegImm(const X86EncodingMap* entry, uint8_t reg, int imm) {
913 if (entry->skeleton.prefix1 != 0) {
914 code_buffer_.push_back(entry->skeleton.prefix1);
915 if (entry->skeleton.prefix2 != 0) {
916 code_buffer_.push_back(entry->skeleton.prefix2);
917 }
918 } else {
919 DCHECK_EQ(0, entry->skeleton.prefix2);
920 }
921 if (imm != 1) {
922 code_buffer_.push_back(entry->skeleton.opcode);
923 } else {
924 // Shorter encoding for 1 bit shift
925 code_buffer_.push_back(entry->skeleton.ax_opcode);
926 }
927 if (entry->skeleton.opcode == 0x0F) {
928 code_buffer_.push_back(entry->skeleton.extra_opcode1);
929 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
930 code_buffer_.push_back(entry->skeleton.extra_opcode2);
931 } else {
932 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
933 }
934 } else {
935 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
936 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
937 }
938 if (reg >= 4) {
939 DCHECK(strchr(entry->name, '8') == NULL) << entry->name << " " << static_cast<int>(reg)
940 << " in " << PrettyMethod(cu_->method_idx, *cu_->dex_file);
941 }
942 DCHECK_LT(reg, 8);
943 uint8_t modrm = (3 << 6) | (entry->skeleton.modrm_opcode << 3) | reg;
944 code_buffer_.push_back(modrm);
945 if (imm != 1) {
946 DCHECK_EQ(entry->skeleton.immediate_bytes, 1);
947 DCHECK(IS_SIMM8(imm));
948 code_buffer_.push_back(imm & 0xFF);
949 }
950 }
951
EmitShiftRegCl(const X86EncodingMap * entry,uint8_t reg,uint8_t cl)952 void X86Mir2Lir::EmitShiftRegCl(const X86EncodingMap* entry, uint8_t reg, uint8_t cl) {
953 DCHECK_EQ(cl, static_cast<uint8_t>(rCX));
954 if (entry->skeleton.prefix1 != 0) {
955 code_buffer_.push_back(entry->skeleton.prefix1);
956 if (entry->skeleton.prefix2 != 0) {
957 code_buffer_.push_back(entry->skeleton.prefix2);
958 }
959 } else {
960 DCHECK_EQ(0, entry->skeleton.prefix2);
961 }
962 code_buffer_.push_back(entry->skeleton.opcode);
963 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
964 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
965 DCHECK_LT(reg, 8);
966 uint8_t modrm = (3 << 6) | (entry->skeleton.modrm_opcode << 3) | reg;
967 code_buffer_.push_back(modrm);
968 DCHECK_EQ(0, entry->skeleton.ax_opcode);
969 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
970 }
971
EmitRegCond(const X86EncodingMap * entry,uint8_t reg,uint8_t condition)972 void X86Mir2Lir::EmitRegCond(const X86EncodingMap* entry, uint8_t reg, uint8_t condition) {
973 if (entry->skeleton.prefix1 != 0) {
974 code_buffer_.push_back(entry->skeleton.prefix1);
975 if (entry->skeleton.prefix2 != 0) {
976 code_buffer_.push_back(entry->skeleton.prefix2);
977 }
978 } else {
979 DCHECK_EQ(0, entry->skeleton.prefix2);
980 }
981 DCHECK_EQ(0, entry->skeleton.ax_opcode);
982 DCHECK_EQ(0x0F, entry->skeleton.opcode);
983 code_buffer_.push_back(0x0F);
984 DCHECK_EQ(0x90, entry->skeleton.extra_opcode1);
985 code_buffer_.push_back(0x90 | condition);
986 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
987 DCHECK_LT(reg, 8);
988 uint8_t modrm = (3 << 6) | (entry->skeleton.modrm_opcode << 3) | reg;
989 code_buffer_.push_back(modrm);
990 DCHECK_EQ(entry->skeleton.immediate_bytes, 0);
991 }
992
EmitJmp(const X86EncodingMap * entry,int rel)993 void X86Mir2Lir::EmitJmp(const X86EncodingMap* entry, int rel) {
994 if (entry->opcode == kX86Jmp8) {
995 DCHECK(IS_SIMM8(rel));
996 code_buffer_.push_back(0xEB);
997 code_buffer_.push_back(rel & 0xFF);
998 } else if (entry->opcode == kX86Jmp32) {
999 code_buffer_.push_back(0xE9);
1000 code_buffer_.push_back(rel & 0xFF);
1001 code_buffer_.push_back((rel >> 8) & 0xFF);
1002 code_buffer_.push_back((rel >> 16) & 0xFF);
1003 code_buffer_.push_back((rel >> 24) & 0xFF);
1004 } else {
1005 DCHECK(entry->opcode == kX86JmpR);
1006 code_buffer_.push_back(entry->skeleton.opcode);
1007 uint8_t reg = static_cast<uint8_t>(rel);
1008 DCHECK_LT(reg, 8);
1009 uint8_t modrm = (3 << 6) | (entry->skeleton.modrm_opcode << 3) | reg;
1010 code_buffer_.push_back(modrm);
1011 }
1012 }
1013
EmitJcc(const X86EncodingMap * entry,int rel,uint8_t cc)1014 void X86Mir2Lir::EmitJcc(const X86EncodingMap* entry, int rel, uint8_t cc) {
1015 DCHECK_LT(cc, 16);
1016 if (entry->opcode == kX86Jcc8) {
1017 DCHECK(IS_SIMM8(rel));
1018 code_buffer_.push_back(0x70 | cc);
1019 code_buffer_.push_back(rel & 0xFF);
1020 } else {
1021 DCHECK(entry->opcode == kX86Jcc32);
1022 code_buffer_.push_back(0x0F);
1023 code_buffer_.push_back(0x80 | cc);
1024 code_buffer_.push_back(rel & 0xFF);
1025 code_buffer_.push_back((rel >> 8) & 0xFF);
1026 code_buffer_.push_back((rel >> 16) & 0xFF);
1027 code_buffer_.push_back((rel >> 24) & 0xFF);
1028 }
1029 }
1030
EmitCallMem(const X86EncodingMap * entry,uint8_t base,int disp)1031 void X86Mir2Lir::EmitCallMem(const X86EncodingMap* entry, uint8_t base, int disp) {
1032 if (entry->skeleton.prefix1 != 0) {
1033 code_buffer_.push_back(entry->skeleton.prefix1);
1034 if (entry->skeleton.prefix2 != 0) {
1035 code_buffer_.push_back(entry->skeleton.prefix2);
1036 }
1037 } else {
1038 DCHECK_EQ(0, entry->skeleton.prefix2);
1039 }
1040 code_buffer_.push_back(entry->skeleton.opcode);
1041 if (entry->skeleton.opcode == 0x0F) {
1042 code_buffer_.push_back(entry->skeleton.extra_opcode1);
1043 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
1044 code_buffer_.push_back(entry->skeleton.extra_opcode2);
1045 } else {
1046 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
1047 }
1048 } else {
1049 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
1050 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
1051 }
1052 uint8_t modrm = (ModrmForDisp(base, disp) << 6) | (entry->skeleton.modrm_opcode << 3) | base;
1053 code_buffer_.push_back(modrm);
1054 if (base == rX86_SP) {
1055 // Special SIB for SP base
1056 code_buffer_.push_back(0 << 6 | (rX86_SP << 3) | rX86_SP);
1057 }
1058 EmitDisp(base, disp);
1059 DCHECK_EQ(0, entry->skeleton.ax_opcode);
1060 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
1061 }
1062
EmitCallThread(const X86EncodingMap * entry,int disp)1063 void X86Mir2Lir::EmitCallThread(const X86EncodingMap* entry, int disp) {
1064 DCHECK_NE(entry->skeleton.prefix1, 0);
1065 code_buffer_.push_back(entry->skeleton.prefix1);
1066 if (entry->skeleton.prefix2 != 0) {
1067 code_buffer_.push_back(entry->skeleton.prefix2);
1068 }
1069 code_buffer_.push_back(entry->skeleton.opcode);
1070 if (entry->skeleton.opcode == 0x0F) {
1071 code_buffer_.push_back(entry->skeleton.extra_opcode1);
1072 if (entry->skeleton.extra_opcode1 == 0x38 || entry->skeleton.extra_opcode2 == 0x3A) {
1073 code_buffer_.push_back(entry->skeleton.extra_opcode2);
1074 } else {
1075 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
1076 }
1077 } else {
1078 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
1079 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
1080 }
1081 uint8_t modrm = (0 << 6) | (entry->skeleton.modrm_opcode << 3) | rBP;
1082 code_buffer_.push_back(modrm);
1083 code_buffer_.push_back(disp & 0xFF);
1084 code_buffer_.push_back((disp >> 8) & 0xFF);
1085 code_buffer_.push_back((disp >> 16) & 0xFF);
1086 code_buffer_.push_back((disp >> 24) & 0xFF);
1087 DCHECK_EQ(0, entry->skeleton.ax_opcode);
1088 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
1089 }
1090
EmitPcRel(const X86EncodingMap * entry,uint8_t reg,int base_or_table,uint8_t index,int scale,int table_or_disp)1091 void X86Mir2Lir::EmitPcRel(const X86EncodingMap* entry, uint8_t reg,
1092 int base_or_table, uint8_t index, int scale, int table_or_disp) {
1093 int disp;
1094 if (entry->opcode == kX86PcRelLoadRA) {
1095 Mir2Lir::SwitchTable *tab_rec = reinterpret_cast<Mir2Lir::SwitchTable*>(table_or_disp);
1096 disp = tab_rec->offset;
1097 } else {
1098 DCHECK(entry->opcode == kX86PcRelAdr);
1099 Mir2Lir::FillArrayData *tab_rec = reinterpret_cast<Mir2Lir::FillArrayData*>(base_or_table);
1100 disp = tab_rec->offset;
1101 }
1102 if (entry->skeleton.prefix1 != 0) {
1103 code_buffer_.push_back(entry->skeleton.prefix1);
1104 if (entry->skeleton.prefix2 != 0) {
1105 code_buffer_.push_back(entry->skeleton.prefix2);
1106 }
1107 } else {
1108 DCHECK_EQ(0, entry->skeleton.prefix2);
1109 }
1110 if (X86_FPREG(reg)) {
1111 reg = reg & X86_FP_REG_MASK;
1112 }
1113 DCHECK_LT(reg, 8);
1114 if (entry->opcode == kX86PcRelLoadRA) {
1115 code_buffer_.push_back(entry->skeleton.opcode);
1116 DCHECK_EQ(0, entry->skeleton.extra_opcode1);
1117 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
1118 uint8_t modrm = (2 << 6) | (reg << 3) | rX86_SP;
1119 code_buffer_.push_back(modrm);
1120 DCHECK_LT(scale, 4);
1121 DCHECK_LT(index, 8);
1122 DCHECK_LT(base_or_table, 8);
1123 uint8_t base = static_cast<uint8_t>(base_or_table);
1124 uint8_t sib = (scale << 6) | (index << 3) | base;
1125 code_buffer_.push_back(sib);
1126 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
1127 } else {
1128 code_buffer_.push_back(entry->skeleton.opcode + reg);
1129 }
1130 code_buffer_.push_back(disp & 0xFF);
1131 code_buffer_.push_back((disp >> 8) & 0xFF);
1132 code_buffer_.push_back((disp >> 16) & 0xFF);
1133 code_buffer_.push_back((disp >> 24) & 0xFF);
1134 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
1135 DCHECK_EQ(0, entry->skeleton.ax_opcode);
1136 }
1137
EmitMacro(const X86EncodingMap * entry,uint8_t reg,int offset)1138 void X86Mir2Lir::EmitMacro(const X86EncodingMap* entry, uint8_t reg, int offset) {
1139 DCHECK(entry->opcode == kX86StartOfMethod) << entry->name;
1140 code_buffer_.push_back(0xE8); // call +0
1141 code_buffer_.push_back(0);
1142 code_buffer_.push_back(0);
1143 code_buffer_.push_back(0);
1144 code_buffer_.push_back(0);
1145
1146 DCHECK_LT(reg, 8);
1147 code_buffer_.push_back(0x58 + reg); // pop reg
1148
1149 EmitRegImm(&X86Mir2Lir::EncodingMap[kX86Sub32RI], reg, offset + 5 /* size of call +0 */);
1150 }
1151
EmitUnimplemented(const X86EncodingMap * entry,LIR * lir)1152 void X86Mir2Lir::EmitUnimplemented(const X86EncodingMap* entry, LIR* lir) {
1153 UNIMPLEMENTED(WARNING) << "encoding kind for " << entry->name << " "
1154 << BuildInsnString(entry->fmt, lir, 0);
1155 for (int i = 0; i < GetInsnSize(lir); ++i) {
1156 code_buffer_.push_back(0xCC); // push breakpoint instruction - int 3
1157 }
1158 }
1159
1160 /*
1161 * Assemble the LIR into binary instruction format. Note that we may
1162 * discover that pc-relative displacements may not fit the selected
1163 * instruction. In those cases we will try to substitute a new code
1164 * sequence or request that the trace be shortened and retried.
1165 */
AssembleInstructions(uintptr_t start_addr)1166 AssemblerStatus X86Mir2Lir::AssembleInstructions(uintptr_t start_addr) {
1167 LIR *lir;
1168 AssemblerStatus res = kSuccess; // Assume success
1169
1170 const bool kVerbosePcFixup = false;
1171 for (lir = first_lir_insn_; lir != NULL; lir = NEXT_LIR(lir)) {
1172 if (lir->opcode < 0) {
1173 continue;
1174 }
1175
1176 if (lir->flags.is_nop) {
1177 continue;
1178 }
1179
1180 if (lir->flags.pcRelFixup) {
1181 switch (lir->opcode) {
1182 case kX86Jcc8: {
1183 LIR *target_lir = lir->target;
1184 DCHECK(target_lir != NULL);
1185 int delta = 0;
1186 uintptr_t pc;
1187 if (IS_SIMM8(lir->operands[0])) {
1188 pc = lir->offset + 2 /* opcode + rel8 */;
1189 } else {
1190 pc = lir->offset + 6 /* 2 byte opcode + rel32 */;
1191 }
1192 uintptr_t target = target_lir->offset;
1193 delta = target - pc;
1194 if (IS_SIMM8(delta) != IS_SIMM8(lir->operands[0])) {
1195 if (kVerbosePcFixup) {
1196 LOG(INFO) << "Retry for JCC growth at " << lir->offset
1197 << " delta: " << delta << " old delta: " << lir->operands[0];
1198 }
1199 lir->opcode = kX86Jcc32;
1200 SetupResourceMasks(lir);
1201 res = kRetryAll;
1202 }
1203 if (kVerbosePcFixup) {
1204 LOG(INFO) << "Source:";
1205 DumpLIRInsn(lir, 0);
1206 LOG(INFO) << "Target:";
1207 DumpLIRInsn(target_lir, 0);
1208 LOG(INFO) << "Delta " << delta;
1209 }
1210 lir->operands[0] = delta;
1211 break;
1212 }
1213 case kX86Jcc32: {
1214 LIR *target_lir = lir->target;
1215 DCHECK(target_lir != NULL);
1216 uintptr_t pc = lir->offset + 6 /* 2 byte opcode + rel32 */;
1217 uintptr_t target = target_lir->offset;
1218 int delta = target - pc;
1219 if (kVerbosePcFixup) {
1220 LOG(INFO) << "Source:";
1221 DumpLIRInsn(lir, 0);
1222 LOG(INFO) << "Target:";
1223 DumpLIRInsn(target_lir, 0);
1224 LOG(INFO) << "Delta " << delta;
1225 }
1226 lir->operands[0] = delta;
1227 break;
1228 }
1229 case kX86Jmp8: {
1230 LIR *target_lir = lir->target;
1231 DCHECK(target_lir != NULL);
1232 int delta = 0;
1233 uintptr_t pc;
1234 if (IS_SIMM8(lir->operands[0])) {
1235 pc = lir->offset + 2 /* opcode + rel8 */;
1236 } else {
1237 pc = lir->offset + 5 /* opcode + rel32 */;
1238 }
1239 uintptr_t target = target_lir->offset;
1240 delta = target - pc;
1241 if (!(cu_->disable_opt & (1 << kSafeOptimizations)) && delta == 0) {
1242 // Useless branch
1243 lir->flags.is_nop = true;
1244 if (kVerbosePcFixup) {
1245 LOG(INFO) << "Retry for useless branch at " << lir->offset;
1246 }
1247 res = kRetryAll;
1248 } else if (IS_SIMM8(delta) != IS_SIMM8(lir->operands[0])) {
1249 if (kVerbosePcFixup) {
1250 LOG(INFO) << "Retry for JMP growth at " << lir->offset;
1251 }
1252 lir->opcode = kX86Jmp32;
1253 SetupResourceMasks(lir);
1254 res = kRetryAll;
1255 }
1256 lir->operands[0] = delta;
1257 break;
1258 }
1259 case kX86Jmp32: {
1260 LIR *target_lir = lir->target;
1261 DCHECK(target_lir != NULL);
1262 uintptr_t pc = lir->offset + 5 /* opcode + rel32 */;
1263 uintptr_t target = target_lir->offset;
1264 int delta = target - pc;
1265 lir->operands[0] = delta;
1266 break;
1267 }
1268 default:
1269 break;
1270 }
1271 }
1272
1273 /*
1274 * If one of the pc-relative instructions expanded we'll have
1275 * to make another pass. Don't bother to fully assemble the
1276 * instruction.
1277 */
1278 if (res != kSuccess) {
1279 continue;
1280 }
1281 CHECK_EQ(static_cast<size_t>(lir->offset), code_buffer_.size());
1282 const X86EncodingMap *entry = &X86Mir2Lir::EncodingMap[lir->opcode];
1283 size_t starting_cbuf_size = code_buffer_.size();
1284 switch (entry->kind) {
1285 case kData: // 4 bytes of data
1286 code_buffer_.push_back(lir->operands[0]);
1287 break;
1288 case kNullary: // 1 byte of opcode
1289 DCHECK_EQ(0, entry->skeleton.prefix1);
1290 DCHECK_EQ(0, entry->skeleton.prefix2);
1291 code_buffer_.push_back(entry->skeleton.opcode);
1292 if (entry->skeleton.extra_opcode1 != 0) {
1293 code_buffer_.push_back(entry->skeleton.extra_opcode1);
1294 if (entry->skeleton.extra_opcode2 != 0) {
1295 code_buffer_.push_back(entry->skeleton.extra_opcode2);
1296 }
1297 } else {
1298 DCHECK_EQ(0, entry->skeleton.extra_opcode2);
1299 }
1300 DCHECK_EQ(0, entry->skeleton.modrm_opcode);
1301 DCHECK_EQ(0, entry->skeleton.ax_opcode);
1302 DCHECK_EQ(0, entry->skeleton.immediate_bytes);
1303 break;
1304 case kReg: // lir operands - 0: reg
1305 EmitOpReg(entry, lir->operands[0]);
1306 break;
1307 case kMem: // lir operands - 0: base, 1: disp
1308 EmitOpMem(entry, lir->operands[0], lir->operands[1]);
1309 break;
1310 case kMemReg: // lir operands - 0: base, 1: disp, 2: reg
1311 EmitMemReg(entry, lir->operands[0], lir->operands[1], lir->operands[2]);
1312 break;
1313 case kArrayReg: // lir operands - 0: base, 1: index, 2: scale, 3: disp, 4: reg
1314 EmitArrayReg(entry, lir->operands[0], lir->operands[1], lir->operands[2],
1315 lir->operands[3], lir->operands[4]);
1316 break;
1317 case kRegMem: // lir operands - 0: reg, 1: base, 2: disp
1318 EmitRegMem(entry, lir->operands[0], lir->operands[1], lir->operands[2]);
1319 break;
1320 case kRegArray: // lir operands - 0: reg, 1: base, 2: index, 3: scale, 4: disp
1321 EmitRegArray(entry, lir->operands[0], lir->operands[1], lir->operands[2],
1322 lir->operands[3], lir->operands[4]);
1323 break;
1324 case kRegThread: // lir operands - 0: reg, 1: disp
1325 EmitRegThread(entry, lir->operands[0], lir->operands[1]);
1326 break;
1327 case kRegReg: // lir operands - 0: reg1, 1: reg2
1328 EmitRegReg(entry, lir->operands[0], lir->operands[1]);
1329 break;
1330 case kRegRegStore: // lir operands - 0: reg2, 1: reg1
1331 EmitRegReg(entry, lir->operands[1], lir->operands[0]);
1332 break;
1333 case kRegRegImm:
1334 EmitRegRegImm(entry, lir->operands[0], lir->operands[1], lir->operands[2]);
1335 break;
1336 case kRegImm: // lir operands - 0: reg, 1: immediate
1337 EmitRegImm(entry, lir->operands[0], lir->operands[1]);
1338 break;
1339 case kThreadImm: // lir operands - 0: disp, 1: immediate
1340 EmitThreadImm(entry, lir->operands[0], lir->operands[1]);
1341 break;
1342 case kMovRegImm: // lir operands - 0: reg, 1: immediate
1343 EmitMovRegImm(entry, lir->operands[0], lir->operands[1]);
1344 break;
1345 case kShiftRegImm: // lir operands - 0: reg, 1: immediate
1346 EmitShiftRegImm(entry, lir->operands[0], lir->operands[1]);
1347 break;
1348 case kShiftRegCl: // lir operands - 0: reg, 1: cl
1349 EmitShiftRegCl(entry, lir->operands[0], lir->operands[1]);
1350 break;
1351 case kRegCond: // lir operands - 0: reg, 1: condition
1352 EmitRegCond(entry, lir->operands[0], lir->operands[1]);
1353 break;
1354 case kJmp: // lir operands - 0: rel
1355 if (entry->opcode == kX86JmpT) {
1356 // This works since the instruction format for jmp and call is basically the same and
1357 // EmitCallThread loads opcode info.
1358 EmitCallThread(entry, lir->operands[0]);
1359 } else {
1360 EmitJmp(entry, lir->operands[0]);
1361 }
1362 break;
1363 case kJcc: // lir operands - 0: rel, 1: CC, target assigned
1364 EmitJcc(entry, lir->operands[0], lir->operands[1]);
1365 break;
1366 case kCall:
1367 switch (entry->opcode) {
1368 case kX86CallM: // lir operands - 0: base, 1: disp
1369 EmitCallMem(entry, lir->operands[0], lir->operands[1]);
1370 break;
1371 case kX86CallT: // lir operands - 0: disp
1372 EmitCallThread(entry, lir->operands[0]);
1373 break;
1374 default:
1375 EmitUnimplemented(entry, lir);
1376 break;
1377 }
1378 break;
1379 case kPcRel: // lir operands - 0: reg, 1: base, 2: index, 3: scale, 4: table
1380 EmitPcRel(entry, lir->operands[0], lir->operands[1], lir->operands[2],
1381 lir->operands[3], lir->operands[4]);
1382 break;
1383 case kMacro:
1384 EmitMacro(entry, lir->operands[0], lir->offset);
1385 break;
1386 default:
1387 EmitUnimplemented(entry, lir);
1388 break;
1389 }
1390 CHECK_EQ(static_cast<size_t>(GetInsnSize(lir)),
1391 code_buffer_.size() - starting_cbuf_size)
1392 << "Instruction size mismatch for entry: " << X86Mir2Lir::EncodingMap[lir->opcode].name;
1393 }
1394 return res;
1395 }
1396
1397 } // namespace art
1398