mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-01 00:12:50 +01:00
4a589eb424
32-bit writes on amd64 zero out the high bits of the corresponding 64-bit register. LLVM makes use of this for zero-extension, but until now relied on custom MCLowering and other code to fixup instructions. Now we have proper handling of sub-registers, this can be done by creating SUBREG_TO_REG instructions at selection-time. Should be no change in functionality. llvm-svn: 182921
173 lines
8.6 KiB
TableGen
173 lines
8.6 KiB
TableGen
//===-- X86InstrExtension.td - Sign and Zero Extensions ----*- tablegen -*-===//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file is distributed under the University of Illinois Open Source
|
|
// License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
// This file describes the sign and zero extension operations.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
let neverHasSideEffects = 1 in {
|
|
let Defs = [AX], Uses = [AL] in
|
|
def CBW : I<0x98, RawFrm, (outs), (ins),
|
|
"{cbtw|cbw}", []>, OpSize; // AX = signext(AL)
|
|
let Defs = [EAX], Uses = [AX] in
|
|
def CWDE : I<0x98, RawFrm, (outs), (ins),
|
|
"{cwtl|cwde}", []>; // EAX = signext(AX)
|
|
|
|
let Defs = [AX,DX], Uses = [AX] in
|
|
def CWD : I<0x99, RawFrm, (outs), (ins),
|
|
"{cwtd|cwd}", []>, OpSize; // DX:AX = signext(AX)
|
|
let Defs = [EAX,EDX], Uses = [EAX] in
|
|
def CDQ : I<0x99, RawFrm, (outs), (ins),
|
|
"{cltd|cdq}", []>; // EDX:EAX = signext(EAX)
|
|
|
|
|
|
let Defs = [RAX], Uses = [EAX] in
|
|
def CDQE : RI<0x98, RawFrm, (outs), (ins),
|
|
"{cltq|cdqe}", []>; // RAX = signext(EAX)
|
|
|
|
let Defs = [RAX,RDX], Uses = [RAX] in
|
|
def CQO : RI<0x99, RawFrm, (outs), (ins),
|
|
"{cqto|cqo}", []>; // RDX:RAX = signext(RAX)
|
|
}
|
|
|
|
|
|
|
|
// Sign/Zero extenders
|
|
let neverHasSideEffects = 1 in {
|
|
def MOVSX16rr8 : I<0xBE, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
|
|
"movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_R8>,
|
|
TB, OpSize, Sched<[WriteALU]>;
|
|
let mayLoad = 1 in
|
|
def MOVSX16rm8 : I<0xBE, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
|
|
"movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_M8>,
|
|
TB, OpSize, Sched<[WriteALULd]>;
|
|
} // neverHasSideEffects = 1
|
|
def MOVSX32rr8 : I<0xBE, MRMSrcReg, (outs GR32:$dst), (ins GR8:$src),
|
|
"movs{bl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB,
|
|
Sched<[WriteALU]>;
|
|
def MOVSX32rm8 : I<0xBE, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
|
|
"movs{bl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB,
|
|
Sched<[WriteALULd]>;
|
|
def MOVSX32rr16: I<0xBF, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
|
|
"movs{wl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB,
|
|
Sched<[WriteALU]>;
|
|
def MOVSX32rm16: I<0xBF, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
|
|
"movs{wl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (sextloadi32i16 addr:$src))], IIC_MOVSX>,
|
|
TB, Sched<[WriteALULd]>;
|
|
|
|
let neverHasSideEffects = 1 in {
|
|
def MOVZX16rr8 : I<0xB6, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
|
|
"movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_R8>,
|
|
TB, OpSize, Sched<[WriteALU]>;
|
|
let mayLoad = 1 in
|
|
def MOVZX16rm8 : I<0xB6, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
|
|
"movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_M8>,
|
|
TB, OpSize, Sched<[WriteALULd]>;
|
|
} // neverHasSideEffects = 1
|
|
def MOVZX32rr8 : I<0xB6, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src),
|
|
"movz{bl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB,
|
|
Sched<[WriteALU]>;
|
|
def MOVZX32rm8 : I<0xB6, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
|
|
"movz{bl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB,
|
|
Sched<[WriteALULd]>;
|
|
def MOVZX32rr16: I<0xB7, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
|
|
"movz{wl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB,
|
|
Sched<[WriteALU]>;
|
|
def MOVZX32rm16: I<0xB7, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
|
|
"movz{wl|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR32:$dst, (zextloadi32i16 addr:$src))], IIC_MOVZX>,
|
|
TB, Sched<[WriteALULd]>;
|
|
|
|
// These are the same as the regular MOVZX32rr8 and MOVZX32rm8
|
|
// except that they use GR32_NOREX for the output operand register class
|
|
// instead of GR32. This allows them to operate on h registers on x86-64.
|
|
let neverHasSideEffects = 1, isCodeGenOnly = 1 in {
|
|
def MOVZX32_NOREXrr8 : I<0xB6, MRMSrcReg,
|
|
(outs GR32_NOREX:$dst), (ins GR8_NOREX:$src),
|
|
"movz{bl|x}\t{$src, $dst|$dst, $src}",
|
|
[], IIC_MOVZX>, TB, Sched<[WriteALU]>;
|
|
let mayLoad = 1 in
|
|
def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem,
|
|
(outs GR32_NOREX:$dst), (ins i8mem_NOREX:$src),
|
|
"movz{bl|x}\t{$src, $dst|$dst, $src}",
|
|
[], IIC_MOVZX>, TB, Sched<[WriteALULd]>;
|
|
}
|
|
|
|
// MOVSX64rr8 always has a REX prefix and it has an 8-bit register
|
|
// operand, which makes it a rare instruction with an 8-bit register
|
|
// operand that can never access an h register. If support for h registers
|
|
// were generalized, this would require a special register class.
|
|
def MOVSX64rr8 : RI<0xBE, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
|
|
"movs{bq|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB,
|
|
Sched<[WriteALU]>;
|
|
def MOVSX64rm8 : RI<0xBE, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
|
|
"movs{bq|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR64:$dst, (sextloadi64i8 addr:$src))], IIC_MOVSX>,
|
|
TB, Sched<[WriteALULd]>;
|
|
def MOVSX64rr16: RI<0xBF, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
|
|
"movs{wq|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB,
|
|
Sched<[WriteALU]>;
|
|
def MOVSX64rm16: RI<0xBF, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
|
|
"movs{wq|x}\t{$src, $dst|$dst, $src}",
|
|
[(set GR64:$dst, (sextloadi64i16 addr:$src))], IIC_MOVSX>,
|
|
TB, Sched<[WriteALULd]>;
|
|
def MOVSX64rr32: RI<0x63, MRMSrcReg, (outs GR64:$dst), (ins GR32:$src),
|
|
"movs{lq|xd}\t{$src, $dst|$dst, $src}",
|
|
[(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>,
|
|
Sched<[WriteALU]>;
|
|
def MOVSX64rm32: RI<0x63, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
|
|
"movs{lq|xd}\t{$src, $dst|$dst, $src}",
|
|
[(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>,
|
|
Sched<[WriteALULd]>;
|
|
|
|
// movzbq and movzwq encodings for the disassembler
|
|
def MOVZX64rr8_Q : RI<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8:$src),
|
|
"movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
|
|
TB, Sched<[WriteALU]>;
|
|
def MOVZX64rm8_Q : RI<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem:$src),
|
|
"movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
|
|
TB, Sched<[WriteALULd]>;
|
|
def MOVZX64rr16_Q : RI<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
|
|
"movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
|
|
TB, Sched<[WriteALU]>;
|
|
def MOVZX64rm16_Q : RI<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
|
|
"movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
|
|
TB, Sched<[WriteALULd]>;
|
|
|
|
// 64-bit zero-extension patterns use SUBREG_TO_REG and an operation writing a
|
|
// 32-bit register.
|
|
def : Pat<(i64 (zext GR8:$src)),
|
|
(SUBREG_TO_REG (i64 0), (MOVZX32rr8 GR8:$src), sub_32bit)>;
|
|
def : Pat<(zextloadi64i8 addr:$src),
|
|
(SUBREG_TO_REG (i64 0), (MOVZX32rm8 addr:$src), sub_32bit)>;
|
|
|
|
def : Pat<(i64 (zext GR16:$src)),
|
|
(SUBREG_TO_REG (i64 0), (MOVZX32rr16 GR16:$src), sub_32bit)>;
|
|
def : Pat<(zextloadi64i16 addr:$src),
|
|
(SUBREG_TO_REG (i64 0), (MOVZX32rm16 addr:$src), sub_32bit)>;
|
|
|
|
// The preferred way to do 32-bit-to-64-bit zero extension on x86-64 is to use a
|
|
// SUBREG_TO_REG to utilize implicit zero-extension, however this isn't possible
|
|
// when the 32-bit value is defined by a truncate or is copied from something
|
|
// where the high bits aren't necessarily all zero. In such cases, we fall back
|
|
// to these explicit zext instructions.
|
|
def : Pat<(i64 (zext GR32:$src)),
|
|
(SUBREG_TO_REG (i64 0), (MOV32rr GR32:$src), sub_32bit)>;
|
|
def : Pat<(i64 (zextloadi64i32 addr:$src)),
|
|
(SUBREG_TO_REG (i64 0), (MOV32rm addr:$src), sub_32bit)>;
|