1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
//! Encodes VEX instructions. These instructions are those added by the Advanced Vector Extensions
//! (AVX).
use super::evex::Register;
use super::rex::{LegacyPrefixes, OpcodeMap};
use super::ByteSink;
use crate::ir::TrapCode;
use crate::isa::x64::args::Amode;
use crate::isa::x64::encoding::rex;
use crate::isa::x64::inst::Inst;
use crate::machinst::MachBuffer;
/// Constructs a VEX-encoded instruction using a builder pattern. This approach makes it visually
/// easier to transform something the manual's syntax, `VEX.128.66.0F 73 /7 ib` to code:
/// `VexInstruction::new().length(...).prefix(...).map(...).w(true).opcode(0x1F).reg(...).rm(...)`.
pub struct VexInstruction {
length: VexVectorLength,
prefix: LegacyPrefixes,
map: OpcodeMap,
opcode: u8,
w: bool,
reg: u8,
rm: RegisterOrAmode,
vvvv: Option<Register>,
imm: Option<u8>,
}
#[allow(missing_docs)]
pub enum RegisterOrAmode {
Register(Register),
Amode(Amode),
}
impl From<u8> for RegisterOrAmode {
fn from(reg: u8) -> Self {
RegisterOrAmode::Register(reg.into())
}
}
impl From<Amode> for RegisterOrAmode {
fn from(amode: Amode) -> Self {
RegisterOrAmode::Amode(amode)
}
}
impl Default for VexInstruction {
fn default() -> Self {
Self {
length: VexVectorLength::default(),
prefix: LegacyPrefixes::None,
map: OpcodeMap::None,
opcode: 0x00,
w: false,
reg: 0x00,
rm: RegisterOrAmode::Register(Register::default()),
vvvv: None,
imm: None,
}
}
}
impl VexInstruction {
/// Construct a default VEX instruction.
pub fn new() -> Self {
Self::default()
}
/// Set the length of the instruction.
#[inline(always)]
pub fn length(mut self, length: VexVectorLength) -> Self {
self.length = length;
self
}
/// Set the legacy prefix byte of the instruction: None | 66 | F2 | F3. VEX instructions
/// pack these into the prefix, not as separate bytes.
#[inline(always)]
pub fn prefix(mut self, prefix: LegacyPrefixes) -> Self {
debug_assert!(
prefix == LegacyPrefixes::None
|| prefix == LegacyPrefixes::_66
|| prefix == LegacyPrefixes::_F2
|| prefix == LegacyPrefixes::_F3
);
self.prefix = prefix;
self
}
/// Set the opcode map byte of the instruction: None | 0F | 0F38 | 0F3A. VEX instructions pack
/// these into the prefix, not as separate bytes.
#[inline(always)]
pub fn map(mut self, map: OpcodeMap) -> Self {
self.map = map;
self
}
/// Set the W bit, denoted by `.W1` or `.W0` in the instruction string.
/// Typically used to indicate an instruction using 64 bits of an operand (e.g.
/// 64 bit lanes). EVEX packs this bit in the EVEX prefix; previous encodings used the REX
/// prefix.
#[inline(always)]
pub fn w(mut self, w: bool) -> Self {
self.w = w;
self
}
/// Set the instruction opcode byte.
#[inline(always)]
pub fn opcode(mut self, opcode: u8) -> Self {
self.opcode = opcode;
self
}
/// Set the register to use for the `reg` bits; many instructions use this as the write operand.
#[inline(always)]
pub fn reg(mut self, reg: impl Into<Register>) -> Self {
self.reg = reg.into().into();
self
}
/// Some instructions use the ModRM.reg field as an opcode extension. This is usually denoted by
/// a `/n` field in the manual.
#[inline(always)]
pub fn opcode_ext(mut self, n: u8) -> Self {
self.reg = n;
self
}
/// Set the register to use for the `rm` bits; many instructions use this
/// as the "read from register/memory" operand. Setting this affects both
/// the ModRM byte (`rm` section) and the VEX prefix (the extension bits
/// for register encodings > 8).
#[inline(always)]
pub fn rm(mut self, reg: impl Into<RegisterOrAmode>) -> Self {
self.rm = reg.into();
self
}
/// Set the `vvvv` register; some instructions allow using this as a second, non-destructive
/// source register in 3-operand instructions (e.g. 2 read, 1 write).
#[allow(dead_code)]
#[inline(always)]
pub fn vvvv(mut self, reg: impl Into<Register>) -> Self {
self.vvvv = Some(reg.into());
self
}
/// Set the imm byte when used for a register. The reg bits are stored in `imm8[7:4]` with
/// the lower bits unused. Overrides a previously set [Self::imm] field.
#[inline(always)]
pub fn imm_reg(mut self, reg: impl Into<Register>) -> Self {
let reg: u8 = reg.into().into();
self.imm = Some((reg & 0xf) << 4);
self
}
/// Set the imm byte.
/// Overrides a previously set [Self::imm_reg] field.
#[inline(always)]
pub fn imm(mut self, imm: u8) -> Self {
self.imm = Some(imm);
self
}
/// The R bit in encoded format (inverted).
#[inline(always)]
fn r_bit(&self) -> u8 {
(!(self.reg >> 3)) & 1
}
/// The X bit in encoded format (inverted).
#[inline(always)]
fn x_bit(&self) -> u8 {
let reg = match &self.rm {
RegisterOrAmode::Register(_) => 0,
RegisterOrAmode::Amode(Amode::ImmReg { .. }) => 0,
RegisterOrAmode::Amode(Amode::ImmRegRegShift { index, .. }) => {
index.to_real_reg().unwrap().hw_enc()
}
RegisterOrAmode::Amode(Amode::RipRelative { .. }) => 0,
};
!(reg >> 3) & 1
}
/// The B bit in encoded format (inverted).
#[inline(always)]
fn b_bit(&self) -> u8 {
let reg = match &self.rm {
RegisterOrAmode::Register(r) => (*r).into(),
RegisterOrAmode::Amode(Amode::ImmReg { base, .. }) => {
base.to_real_reg().unwrap().hw_enc()
}
RegisterOrAmode::Amode(Amode::ImmRegRegShift { base, .. }) => {
base.to_real_reg().unwrap().hw_enc()
}
RegisterOrAmode::Amode(Amode::RipRelative { .. }) => 0,
};
!(reg >> 3) & 1
}
/// Is the 2 byte prefix available for this instruction?
/// We essentially just check if we need any of the bits that are only available
/// in the 3 byte instruction
#[inline(always)]
fn use_2byte_prefix(&self) -> bool {
// These bits are only represented on the 3 byte prefix, so their presence
// implies the use of the 3 byte prefix
self.b_bit() == 1 && self.x_bit() == 1 &&
// The presence of W1 in the opcode column implies the opcode must be encoded using the
// 3-byte form of the VEX prefix.
self.w == false &&
// The presence of 0F3A and 0F38 in the opcode column implies that opcode can only be
// encoded by the three-byte form of VEX
!(self.map == OpcodeMap::_0F3A || self.map == OpcodeMap::_0F38)
}
/// The last byte of the 2byte and 3byte prefixes is mostly the same, share the common
/// encoding logic here.
#[inline(always)]
fn prefix_last_byte(&self) -> u8 {
let vvvv = self.vvvv.map(|r| r.into()).unwrap_or(0x00);
let mut byte = 0x00;
byte |= self.prefix.bits();
byte |= self.length.bits() << 2;
byte |= ((!vvvv) & 0xF) << 3;
byte
}
/// Encode the 2 byte prefix
#[inline(always)]
fn encode_2byte_prefix<CS: ByteSink + ?Sized>(&self, sink: &mut CS) {
// 2 bytes:
// +-----+ +-------------------+
// | C5h | | R | vvvv | L | pp |
// +-----+ +-------------------+
let last_byte = self.prefix_last_byte() | (self.r_bit() << 7);
sink.put1(0xC5);
sink.put1(last_byte);
}
/// Encode the 3 byte prefix
#[inline(always)]
fn encode_3byte_prefix<CS: ByteSink + ?Sized>(&self, sink: &mut CS) {
// 3 bytes:
// +-----+ +--------------+ +-------------------+
// | C4h | | RXB | m-mmmm | | W | vvvv | L | pp |
// +-----+ +--------------+ +-------------------+
let mut second_byte = 0x00;
second_byte |= self.map.bits(); // m-mmmm field
second_byte |= self.b_bit() << 5;
second_byte |= self.x_bit() << 6;
second_byte |= self.r_bit() << 7;
let w_bit = self.w as u8;
let last_byte = self.prefix_last_byte() | (w_bit << 7);
sink.put1(0xC4);
sink.put1(second_byte);
sink.put1(last_byte);
}
/// Emit the VEX-encoded instruction to the provided buffer.
pub fn encode(&self, sink: &mut MachBuffer<Inst>) {
if let RegisterOrAmode::Amode(amode) = &self.rm {
if amode.can_trap() {
sink.add_trap(TrapCode::HeapOutOfBounds);
}
}
// 2/3 byte prefix
if self.use_2byte_prefix() {
self.encode_2byte_prefix(sink);
} else {
self.encode_3byte_prefix(sink);
}
// 1 Byte Opcode
sink.put1(self.opcode);
match &self.rm {
// Not all instructions use Reg as a reg, some use it as an extension
// of the opcode.
RegisterOrAmode::Register(reg) => {
let rm: u8 = (*reg).into();
sink.put1(rex::encode_modrm(3, self.reg & 7, rm & 7));
}
// For address-based modes reuse the logic from the `rex` module
// for the modrm and trailing bytes since VEX uses the same
// encoding.
RegisterOrAmode::Amode(amode) => {
let bytes_at_end = if self.imm.is_some() { 1 } else { 0 };
rex::emit_modrm_sib_disp(sink, self.reg & 7, amode, bytes_at_end);
}
}
// Optional 1 Byte imm
if let Some(imm) = self.imm {
sink.put1(imm);
}
}
}
/// The VEX format allows choosing a vector length in the `L` bit.
#[allow(dead_code, missing_docs)] // Wider-length vectors are not yet used.
pub enum VexVectorLength {
V128,
V256,
}
impl VexVectorLength {
/// Encode the `L` bit.
fn bits(&self) -> u8 {
match self {
Self::V128 => 0b0,
Self::V256 => 0b1,
}
}
}
impl Default for VexVectorLength {
fn default() -> Self {
Self::V128
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::isa::x64::inst::args::Gpr;
use crate::isa::x64::inst::regs;
use crate::opts::MemFlags;
#[test]
fn vpslldq() {
// VEX.128.66.0F 73 /7 ib
// VPSLLDQ xmm1, xmm2, imm8
let dst = regs::xmm1().to_real_reg().unwrap().hw_enc();
let src = regs::xmm2().to_real_reg().unwrap().hw_enc();
let mut sink = MachBuffer::new();
VexInstruction::new()
.length(VexVectorLength::V128)
.prefix(LegacyPrefixes::_66)
.map(OpcodeMap::_0F)
.opcode(0x73)
.opcode_ext(7)
.vvvv(dst)
.rm(src)
.imm(0x17)
.encode(&mut sink);
let bytes = sink.finish().data;
assert_eq!(bytes.as_slice(), [0xc5, 0xf1, 0x73, 0xfa, 0x17]);
}
#[test]
fn vblendvpd() {
// A four operand instruction
// VEX.128.66.0F3A.W0 4B /r /is4
// VBLENDVPD xmm1, xmm2, xmm3, xmm4
let dst = regs::xmm1().to_real_reg().unwrap().hw_enc();
let a = regs::xmm2().to_real_reg().unwrap().hw_enc();
let b = regs::xmm3().to_real_reg().unwrap().hw_enc();
let c = regs::xmm4().to_real_reg().unwrap().hw_enc();
let mut sink = MachBuffer::new();
VexInstruction::new()
.length(VexVectorLength::V128)
.prefix(LegacyPrefixes::_66)
.map(OpcodeMap::_0F3A)
.w(false)
.opcode(0x4B)
.reg(dst)
.vvvv(a)
.rm(b)
.imm_reg(c)
.encode(&mut sink);
let bytes = sink.finish().data;
assert_eq!(bytes.as_slice(), [0xc4, 0xe3, 0x69, 0x4b, 0xcb, 0x40]);
}
#[test]
fn vcmpps() {
// VEX.128.0F.WIG C2 /r ib
// VCMPPS ymm10, ymm11, ymm12, 4 // neq
let dst = regs::xmm10().to_real_reg().unwrap().hw_enc();
let a = regs::xmm11().to_real_reg().unwrap().hw_enc();
let b = regs::xmm12().to_real_reg().unwrap().hw_enc();
let mut sink = MachBuffer::new();
VexInstruction::new()
.length(VexVectorLength::V256)
.prefix(LegacyPrefixes::None)
.map(OpcodeMap::_0F)
.opcode(0xC2)
.reg(dst)
.vvvv(a)
.rm(b)
.imm(4)
.encode(&mut sink);
let bytes = sink.finish().data;
assert_eq!(bytes.as_slice(), [0xc4, 0x41, 0x24, 0xc2, 0xd4, 0x04]);
}
#[test]
fn vandnps() {
// VEX.128.0F 55 /r
// VANDNPS xmm0, xmm1, xmm2
let dst = regs::xmm2().to_real_reg().unwrap().hw_enc();
let src1 = regs::xmm1().to_real_reg().unwrap().hw_enc();
let src2 = regs::xmm0().to_real_reg().unwrap().hw_enc();
let mut sink = MachBuffer::new();
VexInstruction::new()
.length(VexVectorLength::V128)
.prefix(LegacyPrefixes::None)
.map(OpcodeMap::_0F)
.opcode(0x55)
.reg(dst)
.vvvv(src1)
.rm(src2)
.encode(&mut sink);
let bytes = sink.finish().data;
assert_eq!(bytes.as_slice(), [0xc5, 0xf0, 0x55, 0xd0]);
}
#[test]
fn vandnps_mem() {
// VEX.128.0F 55 /r
// VANDNPS 10(%r13), xmm1, xmm2
let dst = regs::xmm2().to_real_reg().unwrap().hw_enc();
let src1 = regs::xmm1().to_real_reg().unwrap().hw_enc();
let src2 = Amode::ImmReg {
base: regs::r13(),
flags: MemFlags::trusted(),
simm32: 10,
};
let mut sink = MachBuffer::new();
VexInstruction::new()
.length(VexVectorLength::V128)
.prefix(LegacyPrefixes::None)
.map(OpcodeMap::_0F)
.opcode(0x55)
.reg(dst)
.vvvv(src1)
.rm(src2)
.encode(&mut sink);
let bytes = sink.finish().data;
assert_eq!(bytes.as_slice(), [0xc4, 0xc1, 0x70, 0x55, 0x55, 0x0a]);
}
#[test]
fn vandnps_more_mem() {
// VEX.128.0F 55 /r
// VANDNPS 100(%rax,%r13,4), xmm1, xmm2
let dst = regs::xmm2().to_real_reg().unwrap().hw_enc();
let src1 = regs::xmm1().to_real_reg().unwrap().hw_enc();
let src2 = Amode::ImmRegRegShift {
base: Gpr::new(regs::rax()).unwrap(),
index: Gpr::new(regs::r13()).unwrap(),
flags: MemFlags::trusted(),
simm32: 100,
shift: 2,
};
let mut sink = MachBuffer::new();
VexInstruction::new()
.length(VexVectorLength::V128)
.prefix(LegacyPrefixes::None)
.map(OpcodeMap::_0F)
.opcode(0x55)
.reg(dst)
.vvvv(src1)
.rm(src2)
.encode(&mut sink);
let bytes = sink.finish().data;
assert_eq!(bytes.as_slice(), [0xc4, 0xa1, 0x70, 0x55, 0x54, 0xa8, 100]);
}
}