1//===-- X86InstrExtension.td - Sign and Zero Extensions ----*- tablegen -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file describes the sign and zero extension operations.
11//
12//===----------------------------------------------------------------------===//
13
14let neverHasSideEffects = 1 in {
15  let Defs = [AX], Uses = [AL] in
16  def CBW : I<0x98, RawFrm, (outs), (ins),
17              "{cbtw|cbw}", []>, OpSize;   // AX = signext(AL)
18  let Defs = [EAX], Uses = [AX] in
19  def CWDE : I<0x98, RawFrm, (outs), (ins),
20              "{cwtl|cwde}", []>;   // EAX = signext(AX)
21
22  let Defs = [AX,DX], Uses = [AX] in
23  def CWD : I<0x99, RawFrm, (outs), (ins),
24              "{cwtd|cwd}", []>, OpSize; // DX:AX = signext(AX)
25  let Defs = [EAX,EDX], Uses = [EAX] in
26  def CDQ : I<0x99, RawFrm, (outs), (ins),
27              "{cltd|cdq}", []>; // EDX:EAX = signext(EAX)
28
29
30  let Defs = [RAX], Uses = [EAX] in
31  def CDQE : RI<0x98, RawFrm, (outs), (ins),
32               "{cltq|cdqe}", []>;     // RAX = signext(EAX)
33
34  let Defs = [RAX,RDX], Uses = [RAX] in
35  def CQO  : RI<0x99, RawFrm, (outs), (ins),
36                "{cqto|cqo}", []>; // RDX:RAX = signext(RAX)
37}
38
39
40
41// Sign/Zero extenders
42let neverHasSideEffects = 1 in {
43def MOVSX16rr8 : I<0xBE, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
44                   "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_R8>,
45                   TB, OpSize, Sched<[WriteALU]>;
46let mayLoad = 1 in
47def MOVSX16rm8 : I<0xBE, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
48                   "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_M8>,
49                   TB, OpSize, Sched<[WriteALULd]>;
50} // neverHasSideEffects = 1
51def MOVSX32rr8 : I<0xBE, MRMSrcReg, (outs GR32:$dst), (ins GR8:$src),
52                   "movs{bl|x}\t{$src, $dst|$dst, $src}",
53                   [(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB,
54                   Sched<[WriteALU]>;
55def MOVSX32rm8 : I<0xBE, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
56                   "movs{bl|x}\t{$src, $dst|$dst, $src}",
57                   [(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB,
58                   Sched<[WriteALULd]>;
59def MOVSX32rr16: I<0xBF, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
60                   "movs{wl|x}\t{$src, $dst|$dst, $src}",
61                   [(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB,
62                   Sched<[WriteALU]>;
63def MOVSX32rm16: I<0xBF, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
64                   "movs{wl|x}\t{$src, $dst|$dst, $src}",
65                   [(set GR32:$dst, (sextloadi32i16 addr:$src))], IIC_MOVSX>,
66                   TB, Sched<[WriteALULd]>;
67
68let neverHasSideEffects = 1 in {
69def MOVZX16rr8 : I<0xB6, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
70                   "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_R8>,
71                   TB, OpSize, Sched<[WriteALU]>;
72let mayLoad = 1 in
73def MOVZX16rm8 : I<0xB6, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
74                   "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_M8>,
75                   TB, OpSize, Sched<[WriteALULd]>;
76} // neverHasSideEffects = 1
77def MOVZX32rr8 : I<0xB6, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src),
78                   "movz{bl|x}\t{$src, $dst|$dst, $src}",
79                   [(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB,
80                   Sched<[WriteALU]>;
81def MOVZX32rm8 : I<0xB6, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
82                   "movz{bl|x}\t{$src, $dst|$dst, $src}",
83                   [(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB,
84                   Sched<[WriteALULd]>;
85def MOVZX32rr16: I<0xB7, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
86                   "movz{wl|x}\t{$src, $dst|$dst, $src}",
87                   [(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB,
88                   Sched<[WriteALU]>;
89def MOVZX32rm16: I<0xB7, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
90                   "movz{wl|x}\t{$src, $dst|$dst, $src}",
91                   [(set GR32:$dst, (zextloadi32i16 addr:$src))], IIC_MOVZX>,
92                   TB, Sched<[WriteALULd]>;
93
94// These are the same as the regular MOVZX32rr8 and MOVZX32rm8
95// except that they use GR32_NOREX for the output operand register class
96// instead of GR32. This allows them to operate on h registers on x86-64.
97let neverHasSideEffects = 1, isCodeGenOnly = 1 in {
98def MOVZX32_NOREXrr8 : I<0xB6, MRMSrcReg,
99                         (outs GR32_NOREX:$dst), (ins GR8_NOREX:$src),
100                         "movz{bl|x}\t{$src, $dst|$dst, $src}",
101                         [], IIC_MOVZX>, TB, Sched<[WriteALU]>;
102let mayLoad = 1 in
103def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem,
104                         (outs GR32_NOREX:$dst), (ins i8mem_NOREX:$src),
105                         "movz{bl|x}\t{$src, $dst|$dst, $src}",
106                         [], IIC_MOVZX>, TB, Sched<[WriteALULd]>;
107}
108
109// MOVSX64rr8 always has a REX prefix and it has an 8-bit register
110// operand, which makes it a rare instruction with an 8-bit register
111// operand that can never access an h register. If support for h registers
112// were generalized, this would require a special register class.
113def MOVSX64rr8 : RI<0xBE, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
114                    "movs{bq|x}\t{$src, $dst|$dst, $src}",
115                    [(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB,
116                    Sched<[WriteALU]>;
117def MOVSX64rm8 : RI<0xBE, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
118                    "movs{bq|x}\t{$src, $dst|$dst, $src}",
119                    [(set GR64:$dst, (sextloadi64i8 addr:$src))], IIC_MOVSX>,
120                    TB, Sched<[WriteALULd]>;
121def MOVSX64rr16: RI<0xBF, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
122                    "movs{wq|x}\t{$src, $dst|$dst, $src}",
123                    [(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB,
124                    Sched<[WriteALU]>;
125def MOVSX64rm16: RI<0xBF, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
126                    "movs{wq|x}\t{$src, $dst|$dst, $src}",
127                    [(set GR64:$dst, (sextloadi64i16 addr:$src))], IIC_MOVSX>,
128                    TB, Sched<[WriteALULd]>;
129def MOVSX64rr32: RI<0x63, MRMSrcReg, (outs GR64:$dst), (ins GR32:$src),
130                    "movs{lq|xd}\t{$src, $dst|$dst, $src}",
131                    [(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>,
132                    Sched<[WriteALU]>;
133def MOVSX64rm32: RI<0x63, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
134                    "movs{lq|xd}\t{$src, $dst|$dst, $src}",
135                    [(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>,
136                    Sched<[WriteALULd]>;
137
138// movzbq and movzwq encodings for the disassembler
139def MOVZX64rr8_Q : RI<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8:$src),
140                       "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
141                       TB, Sched<[WriteALU]>;
142def MOVZX64rm8_Q : RI<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem:$src),
143                       "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
144                       TB, Sched<[WriteALULd]>;
145def MOVZX64rr16_Q : RI<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
146                       "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
147                       TB, Sched<[WriteALU]>;
148def MOVZX64rm16_Q : RI<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
149                       "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
150                       TB, Sched<[WriteALULd]>;
151
152// 64-bit zero-extension patterns use SUBREG_TO_REG and an operation writing a
153// 32-bit register.
154def : Pat<(i64 (zext GR8:$src)),
155          (SUBREG_TO_REG (i64 0), (MOVZX32rr8 GR8:$src), sub_32bit)>;
156def : Pat<(zextloadi64i8 addr:$src),
157          (SUBREG_TO_REG (i64 0), (MOVZX32rm8 addr:$src), sub_32bit)>;
158
159def : Pat<(i64 (zext GR16:$src)),
160          (SUBREG_TO_REG (i64 0), (MOVZX32rr16 GR16:$src), sub_32bit)>;
161def : Pat<(zextloadi64i16 addr:$src),
162          (SUBREG_TO_REG (i64 0), (MOVZX32rm16 addr:$src), sub_32bit)>;
163
164// The preferred way to do 32-bit-to-64-bit zero extension on x86-64 is to use a
165// SUBREG_TO_REG to utilize implicit zero-extension, however this isn't possible
166// when the 32-bit value is defined by a truncate or is copied from something
167// where the high bits aren't necessarily all zero. In such cases, we fall back
168// to these explicit zext instructions.
169def : Pat<(i64 (zext GR32:$src)),
170          (SUBREG_TO_REG (i64 0), (MOV32rr GR32:$src), sub_32bit)>;
171def : Pat<(i64 (zextloadi64i32 addr:$src)),
172          (SUBREG_TO_REG (i64 0), (MOV32rm addr:$src), sub_32bit)>;
173