x64: move encodings to a separate module

In order to benchmark the encoding code with criterion, the functions
and structures must be public. Moving this code to its own module
(instead of keeping as a submodule to `inst`), allows `inst` to remain
private. This avoids having to expose and document (or ignore
documenting) the numerous instruction variants in `inst` while allowing
access to the encoding code. This commit changes no functionality.

cranelift/codegen/src/isa/x64/encoding/evex.rs

@@ -208,6 +208,8 @@ impl EvexInstruction {
     }
 }
 
+/// Describe the register index to use. This wrapper is a type-safe way to pass
+/// around the registers defined in `inst/regs.rs`.
 #[derive(Copy, Clone, Default)]
 pub struct Register(u8);
 impl From<u8> for Register {
@@ -216,13 +218,18 @@ impl From<u8> for Register {
         Self(reg)
     }
 }
+impl Into<u8> for Register {
+    fn into(self) -> u8 {
+        self.0
+    }
+}
 
 /// Defines the EVEX context for the `L'`, `L`, and `b` bits (bits 6:4 of EVEX P2 byte). Table 2-36 in
 /// section 2.6.10 (Intel Software Development Manual, volume 2A) describes how these bits can be
 /// used together for certain classes of instructions; i.e., special care should be taken to ensure
 /// that instructions use an applicable correct `EvexContext`. Table 2-39 contains cases where
 /// opcodes can result in an #UD.
-#[allow(dead_code)] // Rounding and broadcast modes are not yet used.
+#[allow(dead_code, missing_docs)] // Rounding and broadcast modes are not yet used.
 pub enum EvexContext {
     RoundingRegToRegFP {
         rc: EvexRoundingControl,
@@ -250,7 +257,7 @@ impl Default for EvexContext {
 
 impl EvexContext {
     /// Encode the `L'`, `L`, and `b` bits (bits 6:4 of EVEX P2 byte) for merging with the P2 byte.
-    fn bits(&self) -> u8 {
+    pub fn bits(&self) -> u8 {
         match self {
             Self::RoundingRegToRegFP { rc } => 0b001 | rc.bits() << 1,
             Self::NoRoundingFP { sae, length } => (*sae as u8) | length.bits() << 1,
@@ -261,7 +268,7 @@ impl EvexContext {
 }
 
 /// The EVEX format allows choosing a vector length in the `L'` and `L` bits; see `EvexContext`.
-#[allow(dead_code)] // Wider-length vectors are not yet used.
+#[allow(dead_code, missing_docs)] // Wider-length vectors are not yet used.
 pub enum EvexVectorLength {
     V128,
     V256,
@@ -287,7 +294,7 @@ impl Default for EvexVectorLength {
 }
 
 /// The EVEX format allows defining rounding control in the `L'` and `L` bits; see `EvexContext`.
-#[allow(dead_code)] // Rounding controls are not yet used.
+#[allow(dead_code, missing_docs)] // Rounding controls are not yet used.
 pub enum EvexRoundingControl {
     RNE,
     RD,
@@ -309,7 +316,7 @@ impl EvexRoundingControl {
 
 /// Defines the EVEX masking behavior; masking support is described in section 2.6.4 of the Intel
 /// Software Development Manual, volume 2A.
-#[allow(dead_code)] // Masking is not yet used.
+#[allow(dead_code, missing_docs)] // Masking is not yet used.
 pub enum EvexMasking {
     None,
     Merging { k: u8 },
@@ -324,7 +331,7 @@ impl Default for EvexMasking {
 
 impl EvexMasking {
     /// Encode the `z` bit for merging with the P2 byte.
-    fn z_bit(&self) -> u8 {
+    pub fn z_bit(&self) -> u8 {
         match self {
             Self::None | Self::Merging { .. } => 0,
             Self::Zeroing { .. } => 1,
@@ -332,7 +339,7 @@ impl EvexMasking {
     }
 
     /// Encode the `aaa` bits for merging with the P2 byte.
-    fn aaa_bits(&self) -> u8 {
+    pub fn aaa_bits(&self) -> u8 {
         match self {
             Self::None => 0b000,
             Self::Merging { k } | Self::Zeroing { k } => {

cranelift/codegen/src/isa/x64/encoding/mod.rs

@@ -1,3 +1,4 @@
+//! Contains the encoding machinery for the various x64 instruction formats.
 use crate::{isa::x64, machinst::MachBuffer};
 use std::vec::Vec;
 
@@ -5,6 +6,8 @@ pub mod evex;
 pub mod rex;
 pub mod vex;
 
+/// The encoding formats in this module all require a way of placing bytes into
+/// a buffer.
 pub trait ByteSink {
     /// Add 1 byte to the code section.
     fn put1(&mut self, _: u8);

cranelift/codegen/src/isa/x64/encoding/rex.rs

@@ -28,8 +28,9 @@ pub(crate) fn low8_will_sign_extend_to_32(x: u32) -> bool {
     xs == ((xs << 24) >> 24)
 }
 
+/// Encode the ModR/M byte.
 #[inline(always)]
-pub(crate) fn encode_modrm(m0d: u8, enc_reg_g: u8, rm_e: u8) -> u8 {
+pub fn encode_modrm(m0d: u8, enc_reg_g: u8, rm_e: u8) -> u8 {
     debug_assert!(m0d < 4);
     debug_assert!(enc_reg_g < 8);
     debug_assert!(rm_e < 8);
@@ -155,6 +156,7 @@ impl From<(OperandSize, Reg)> for RexFlags {
 
 /// Allows using the same opcode byte in different "opcode maps" to allow for more instruction
 /// encodings. See appendix A in the Intel Software Developer's Manual, volume 2A, for more details.
+#[allow(missing_docs)]
 pub enum OpcodeMap {
     None,
     _0F,

cranelift/codegen/src/isa/x64/inst/emit.rs

@@ -2,16 +2,16 @@ use crate::binemit::{Addend, Reloc};
 use crate::ir::immediates::{Ieee32, Ieee64};
 use crate::ir::LibCall;
 use crate::ir::TrapCode;
-use crate::isa::x64::inst::args::*;
+use crate::isa::x64::encoding::evex::{EvexInstruction, EvexVectorLength};
-use crate::isa::x64::inst::*;
+use crate::isa::x64::encoding::rex::{
-use crate::machinst::{inst_common, MachBuffer, MachInstEmit, MachLabel};
-use core::convert::TryInto;
-use encoding::evex::{EvexInstruction, EvexVectorLength};
-use encoding::rex::{
     emit_simm, emit_std_enc_enc, emit_std_enc_mem, emit_std_reg_mem, emit_std_reg_reg, int_reg_enc,
     low8_will_sign_extend_to_32, low8_will_sign_extend_to_64, reg_enc, LegacyPrefixes, OpcodeMap,
     RexFlags,
 };
+use crate::isa::x64::inst::args::*;
+use crate::isa::x64::inst::*;
+use crate::machinst::{inst_common, MachBuffer, MachInstEmit, MachLabel};
+use core::convert::TryInto;
 use log::debug;
 use regalloc::{Reg, Writable};
 

cranelift/codegen/src/isa/x64/inst/mod.rs

@@ -22,7 +22,6 @@ pub mod args;
 mod emit;
 #[cfg(test)]
 mod emit_tests;
-pub(crate) mod encoding;
 pub mod regs;
 pub mod unwind;
 
@@ -2856,7 +2855,7 @@ impl EmitState {
         self.stack_map = None;
     }
 
-    fn cur_srcloc(&self) -> SourceLoc {
+    pub(crate) fn cur_srcloc(&self) -> SourceLoc {
         self.cur_srcloc
     }
 }

cranelift/codegen/src/isa/x64/mod.rs

@@ -18,6 +18,7 @@ use target_lexicon::Triple;
 use crate::isa::unwind::systemv;
 
 mod abi;
+pub mod encoding;
 mod inst;
 mod lower;
 mod settings;