Add a DynRex recipe type for x86, decreasing the number of recipes (#1298)

This patch adds a third mode for templates: REX inference is requestable
at template instantiation time. This reduces the number of recipes
by removing rex()/nonrex() redundancy for many instructions.

cranelift/codegen/meta/src/cdsl/recipes.rs

@@ -172,7 +172,7 @@ pub(crate) struct EncodingRecipeBuilder {
     pub base_size: u64,
     pub operands_in: Option<Vec<OperandConstraint>>,
     pub operands_out: Option<Vec<OperandConstraint>>,
-    compute_size: Option<&'static str>,
+    pub compute_size: Option<&'static str>,
     pub branch_range: Option<BranchRange>,
     pub emit: Option<String>,
     clobbers_flags: Option<bool>,

cranelift/codegen/meta/src/isa/x86/encodings.rs

@@ -140,32 +140,59 @@ impl PerCpuModeEncodings {
         self.enc64.push(encoding);
     }
 
+    /// Adds I32/I64 encodings as appropriate for a typed instruction.
+    /// The REX prefix is always inferred at runtime.
+    ///
     /// Add encodings for `inst.i32` to X86_32.
-    /// Add encodings for `inst.i32` to X86_64 with and without REX.
+    /// Add encodings for `inst.i32` to X86_64 with optional, inferred REX.
     /// Add encodings for `inst.i64` to X86_64 with a REX.W prefix.
     fn enc_i32_i64(&mut self, inst: impl Into<InstSpec>, template: Template) {
         let inst: InstSpec = inst.into();
+
+        // I32 on x86: no REX prefix.
+        self.enc32(inst.bind(I32), template.infer_rex());
+
+        // I32 on x86_64: REX.W unset; REX.RXB determined at runtime from registers.
+        self.enc64(inst.bind(I32), template.infer_rex());
+
+        // I64 on x86_64: REX.W set; REX.RXB determined at runtime from registers.
+        self.enc64(inst.bind(I64), template.infer_rex().w());
+    }
+
+    /// Adds I32/I64 encodings as appropriate for a typed instruction.
+    /// All variants of REX prefix are explicitly emitted, not inferred.
+    ///
+    /// Add encodings for `inst.i32` to X86_32.
+    /// Add encodings for `inst.i32` to X86_64 with and without REX.
+    /// Add encodings for `inst.i64` to X86_64 with and without REX.
+    fn enc_i32_i64_explicit_rex(&mut self, inst: impl Into<InstSpec>, template: Template) {
+        let inst: InstSpec = inst.into();
         self.enc32(inst.bind(I32), template.nonrex());
 
-        // REX-less encoding must come after REX encoding so we don't use it by default. Otherwise
+        // REX-less encoding must come after REX encoding so we don't use it by default.
-        // reg-alloc would never use r8 and up.
+        // Otherwise reg-alloc would never use r8 and up.
         self.enc64(inst.bind(I32), template.rex());
         self.enc64(inst.bind(I32), template.nonrex());
         self.enc64(inst.bind(I64), template.rex().w());
     }
 
-    /// Add encodings for `inst.b32` to X86_32.
+    /// Adds B32/B64 encodings as appropriate for a typed instruction.
-    /// Add encodings for `inst.b32` to X86_64 with and without REX.
+    /// The REX prefix is always inferred at runtime.
-    /// Add encodings for `inst.b64` to X86_64 with a REX.W prefix.
+    ///
+    /// Adds encoding for `inst.b32` to X86_32.
+    /// Adds encoding for `inst.b32` to X86_64 with optional, inferred REX.
+    /// Adds encoding for `inst.b64` to X86_64 with a REX.W prefix.
     fn enc_b32_b64(&mut self, inst: impl Into<InstSpec>, template: Template) {
         let inst: InstSpec = inst.into();
-        self.enc32(inst.bind(B32), template.nonrex());
 
-        // REX-less encoding must come after REX encoding so we don't use it by default. Otherwise
+        // B32 on x86: no REX prefix.
-        // reg-alloc would never use r8 and up.
+        self.enc32(inst.bind(B32), template.infer_rex());
-        self.enc64(inst.bind(B32), template.rex());
+
-        self.enc64(inst.bind(B32), template.nonrex());
+        // B32 on x86_64: REX.W unset; REX.RXB determined at runtime from registers.
-        self.enc64(inst.bind(B64), template.rex().w());
+        self.enc64(inst.bind(B32), template.infer_rex());
+
+        // B64 on x86_64: REX.W set; REX.RXB determined at runtime from registers.
+        self.enc64(inst.bind(B64), template.infer_rex().w());
     }
 
     /// Add encodings for `inst.i32` to X86_32.
@@ -994,8 +1021,8 @@ pub(crate) fn define(
         e.enc_x86_64(istore8.bind(I64).bind(Any), recipe.opcodes(&MOV_BYTE_STORE));
     }
 
-    e.enc_i32_i64(spill, rec_spillSib32.opcodes(&MOV_STORE));
+    e.enc_i32_i64_explicit_rex(spill, rec_spillSib32.opcodes(&MOV_STORE));
-    e.enc_i32_i64(regspill, rec_regspill32.opcodes(&MOV_STORE));
+    e.enc_i32_i64_explicit_rex(regspill, rec_regspill32.opcodes(&MOV_STORE));
     e.enc_r32_r64_rex_only(spill, rec_spillSib32.opcodes(&MOV_STORE));
     e.enc_r32_r64_rex_only(regspill, rec_regspill32.opcodes(&MOV_STORE));
 
@@ -1020,8 +1047,8 @@ pub(crate) fn define(
         e.enc_i32_i64_ld_st(sload8, true, recipe.opcodes(&MOVSX_BYTE));
     }
 
-    e.enc_i32_i64(fill, rec_fillSib32.opcodes(&MOV_LOAD));
+    e.enc_i32_i64_explicit_rex(fill, rec_fillSib32.opcodes(&MOV_LOAD));
-    e.enc_i32_i64(regfill, rec_regfill32.opcodes(&MOV_LOAD));
+    e.enc_i32_i64_explicit_rex(regfill, rec_regfill32.opcodes(&MOV_LOAD));
     e.enc_r32_r64_rex_only(fill, rec_fillSib32.opcodes(&MOV_LOAD));
     e.enc_r32_r64_rex_only(regfill, rec_regfill32.opcodes(&MOV_LOAD));
 

cranelift/codegen/meta/src/isa/x86/recipes.rs

@@ -61,7 +61,7 @@ impl<'builder> RecipeGroup<'builder> {
         self.templates
             .iter()
             .find(|recipe| recipe.name() == name)
-            .unwrap_or_else(|| panic!("unknown tail recipe name: {}. Try recipe?", name))
+            .unwrap_or_else(|| panic!("unknown template name: {}. Try recipe?", name))
     }
 }
 
@@ -132,6 +132,33 @@ fn replace_nonrex_constraints(
         .collect()
 }
 
+/// Specifies how the REX prefix is emitted by a Recipe.
+#[derive(Copy, Clone, PartialEq)]
+pub enum RexRecipeKind {
+    /// The REX emission behavior is not hardcoded for the Recipe
+    /// and may be overridden when using the Template.
+    Unspecified,
+
+    /// The Recipe must hardcode the non-emission of the REX prefix.
+    NeverEmitRex,
+
+    /// The Recipe must hardcode the emission of the REX prefix.
+    AlwaysEmitRex,
+
+    /// The Recipe should infer the emission of the REX.RXB bits from registers,
+    /// and the REX.W bit from the EncodingBits.
+    ///
+    /// Because such a Recipe has a non-constant instruction size, it must have
+    /// a special `compute_size` handler for the inferrable-REX case.
+    InferRex,
+}
+
+impl Default for RexRecipeKind {
+    fn default() -> Self {
+        Self::Unspecified
+    }
+}
+
 /// Previously called a TailRecipe in the Python meta language, this allows to create multiple
 /// variants of a single base EncodingRecipe (rex prefix, specialized w/rrr bits, different
 /// opcodes). It serves as a prototype of an EncodingRecipe, which is then used when actually creating
@@ -145,16 +172,17 @@ pub(crate) struct Template<'builder> {
     /// The recipe template, which is to be specialized (by copy).
     recipe: EncodingRecipeBuilder,
 
-    /// Does this recipe requires a REX prefix?
+    /// How is the REX prefix emitted?
-    requires_prefix: bool,
+    rex_kind: RexRecipeKind,
+
+    /// Function for `compute_size()` when REX is inferrable.
+    inferred_rex_compute_size: Option<&'static str>,
 
     /// Other recipe to use when REX-prefixed.
     when_prefixed: Option<Rc<Template<'builder>>>,
 
-    // Specialized parameters.
+    // Parameters passed in the EncodingBits.
-    /// Should we include the REX prefix?
+    /// Value of the W bit (0 or 1), stored in the EncodingBits.
-    rex: bool,
-    /// Value of the W bit (0 or 1).
     w_bit: u16,
     /// Value of the RRR bits (between 0 and 0b111).
     rrr_bits: u16,
@@ -167,9 +195,9 @@ impl<'builder> Template<'builder> {
         Self {
             regs,
             recipe,
-            requires_prefix: false,
+            rex_kind: RexRecipeKind::default(),
+            inferred_rex_compute_size: None,
             when_prefixed: None,
-            rex: false,
             w_bit: 0,
             rrr_bits: 0,
             op_bytes: &opcodes::EMPTY,
@@ -179,9 +207,15 @@ impl<'builder> Template<'builder> {
     fn name(&self) -> &str {
         &self.recipe.name
     }
-    fn requires_prefix(self, value: bool) -> Self {
+    fn rex_kind(self, kind: RexRecipeKind) -> Self {
         Self {
-            requires_prefix: value,
+            rex_kind: kind,
+            ..self
+        }
+    }
+    fn inferred_rex_compute_size(self, function: &'static str) -> Self {
+        Self {
+            inferred_rex_compute_size: Some(function),
             ..self
         }
     }
@@ -212,12 +246,19 @@ impl<'builder> Template<'builder> {
         copy
     }
     pub fn nonrex(&self) -> Self {
-        assert!(!self.requires_prefix, "Tail recipe requires REX prefix.");
+        assert!(
+            self.rex_kind != RexRecipeKind::AlwaysEmitRex,
+            "Template requires REX prefix."
+        );
         let mut copy = self.clone();
-        copy.rex = false;
+        copy.rex_kind = RexRecipeKind::NeverEmitRex;
         copy
     }
     pub fn rex(&self) -> Self {
+        assert!(
+            self.rex_kind != RexRecipeKind::NeverEmitRex,
+            "Template requires no REX prefix."
+        );
         if let Some(prefixed) = &self.when_prefixed {
             let mut ret = prefixed.rex();
             // Forward specialized parameters.
@@ -227,36 +268,62 @@ impl<'builder> Template<'builder> {
             return ret;
         }
         let mut copy = self.clone();
-        copy.rex = true;
+        copy.rex_kind = RexRecipeKind::AlwaysEmitRex;
+        copy
+    }
+    pub fn infer_rex(&self) -> Self {
+        assert!(
+            self.rex_kind != RexRecipeKind::NeverEmitRex,
+            "Template requires no REX prefix."
+        );
+        assert!(
+            self.when_prefixed.is_none(),
+            "infer_rex used with when_prefixed()."
+        );
+        let mut copy = self.clone();
+        copy.rex_kind = RexRecipeKind::InferRex;
         copy
     }
 
     pub fn build(mut self) -> (EncodingRecipe, u16) {
-        let (name, bits) = decode_opcodes(&self.op_bytes, self.rrr_bits, self.w_bit);
+        let (opcode, bits) = decode_opcodes(&self.op_bytes, self.rrr_bits, self.w_bit);
 
-        let (name, rex_prefix_size) = if self.rex {
+        let (recipe_name, rex_prefix_size) = match self.rex_kind {
-            ("Rex".to_string() + name, 1)
+            RexRecipeKind::Unspecified | RexRecipeKind::NeverEmitRex => {
-        } else {
+                // Ensure the operands are limited to non-REX constraints.
-            (name.into(), 0)
+                let operands_in = self.recipe.operands_in.unwrap_or_default();
+                self.recipe.operands_in = Some(replace_nonrex_constraints(self.regs, operands_in));
+                let operands_out = self.recipe.operands_out.unwrap_or_default();
+                self.recipe.operands_out =
+                    Some(replace_nonrex_constraints(self.regs, operands_out));
+
+                (opcode.into(), 0)
+            }
+            RexRecipeKind::AlwaysEmitRex => ("Rex".to_string() + opcode, 1),
+            RexRecipeKind::InferRex => {
+                // Hook up the right function for inferred compute_size().
+                assert!(
+                    self.inferred_rex_compute_size.is_some(),
+                    "InferRex recipe '{}' needs an inferred_rex_compute_size function.",
+                    &self.recipe.name
+                );
+                self.recipe.compute_size = self.inferred_rex_compute_size;
+
+                ("DynRex".to_string() + opcode, 0)
+            }
         };
 
         let size_addendum = self.op_bytes.len() as u64 + rex_prefix_size;
         self.recipe.base_size += size_addendum;
 
         // Branch ranges are relative to the end of the instruction.
+        // For InferRex, the range should be the minimum, assuming no REX.
         if let Some(range) = self.recipe.branch_range.as_mut() {
             range.inst_size += size_addendum;
         }
 
-        self.recipe.emit = replace_put_op(self.recipe.emit, &name);
+        self.recipe.emit = replace_put_op(self.recipe.emit, &recipe_name);
-        self.recipe.name = name + &self.recipe.name;
+        self.recipe.name = recipe_name + &self.recipe.name;
-
-        if !self.rex {
-            let operands_in = self.recipe.operands_in.unwrap_or_default();
-            self.recipe.operands_in = Some(replace_nonrex_constraints(self.regs, operands_in));
-            let operands_out = self.recipe.operands_out.unwrap_or_default();
-            self.recipe.operands_out = Some(replace_nonrex_constraints(self.regs, operands_out));
-        }
 
         (self.recipe.build(), bits)
     }
@@ -438,7 +505,8 @@ pub(crate) fn define<'shared>(
     );
 
     // XX /r
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("rr", &formats.binary, 1)
                 .operands_in(vec![gpr, gpr])
                 .operands_out(vec![0])
@@ -448,10 +516,14 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, in_reg1, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     // XX /r with operands swapped. (RM form).
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("rrx", &formats.binary, 1)
                 .operands_in(vec![gpr, gpr])
                 .operands_out(vec![0])
@@ -461,6 +533,9 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg1, in_reg0, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     // XX /r with FPR ins and outs. A form.
@@ -513,7 +588,8 @@ pub(crate) fn define<'shared>(
     }
 
     // XX /n for a unary operation with extension bits.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("ur", &formats.unary, 1)
                 .operands_in(vec![gpr])
                 .operands_out(vec![0])
@@ -523,11 +599,15 @@ pub(crate) fn define<'shared>(
                         modrm_r_bits(in_reg0, bits, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
     );
 
     // XX /r, but for a unary operator with separate input/output register, like
     // copies. MR form, preserving flags.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("umr", &formats.unary, 1)
                 .operands_in(vec![gpr])
                 .operands_out(vec![gpr])
@@ -538,6 +618,9 @@ pub(crate) fn define<'shared>(
                         modrm_rr(out_reg0, in_reg0, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_outreg0"),
     );
 
     // Same as umr, but with FPR -> GPR registers.
@@ -643,7 +726,8 @@ pub(crate) fn define<'shared>(
     );
 
     // XX /r, RM form, GPR -> FPR.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("frurm", &formats.unary, 1)
                 .operands_in(vec![gpr])
                 .operands_out(vec![fpr])
@@ -654,6 +738,9 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, out_reg0, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_outreg0"),
     );
 
     // XX /r, RM form, FPR -> GPR.
@@ -734,7 +821,8 @@ pub(crate) fn define<'shared>(
     );
 
     // XX /n for division: inputs in %rax, %rdx, r. Outputs in %rax, %rdx.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("div", &formats.ternary, 1)
                 .operands_in(vec![
                     OperandConstraint::FixedReg(reg_rax),
@@ -749,10 +837,14 @@ pub(crate) fn define<'shared>(
                         modrm_r_bits(in_reg2, bits, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg2"),
     );
 
     // XX /n for {s,u}mulx: inputs in %rax, r. Outputs in %rdx(hi):%rax(lo)
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("mulx", &formats.binary, 1)
                 .operands_in(vec![
                     OperandConstraint::FixedReg(reg_rax),
@@ -768,11 +860,15 @@ pub(crate) fn define<'shared>(
                         modrm_r_bits(in_reg1, bits, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg1"),
     );
 
     // XX /n ib with 8-bit immediate sign-extended.
     {
-        recipes.add_template_recipe(
+        recipes.add_template(
+            Template::new(
                 EncodingRecipeBuilder::new("r_ib", &formats.binary_imm, 2)
                     .operands_in(vec![gpr])
                     .operands_out(vec![0])
@@ -790,6 +886,9 @@ pub(crate) fn define<'shared>(
                             sink.put1(imm as u8);
                         "#,
                     ),
+                regs,
+            )
+            .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
         );
 
         recipes.add_template_recipe(
@@ -813,7 +912,8 @@ pub(crate) fn define<'shared>(
         );
 
         // XX /n id with 32-bit immediate sign-extended.
-        recipes.add_template_recipe(
+        recipes.add_template(
+            Template::new(
                 EncodingRecipeBuilder::new("r_id", &formats.binary_imm, 5)
                     .operands_in(vec![gpr])
                     .operands_out(vec![0])
@@ -831,6 +931,9 @@ pub(crate) fn define<'shared>(
                             sink.put4(imm as u32);
                         "#,
                     ),
+                regs,
+            )
+            .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
         );
     }
 
@@ -1326,7 +1429,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![gpr, gpr])
                 .inst_predicate(has_no_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_or_offset_for_in_reg_1")
+                .compute_size("size_plus_maybe_sib_or_offset_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1354,7 +1457,7 @@ pub(crate) fn define<'shared>(
                     .operands_in(vec![abcd, gpr])
                     .inst_predicate(has_no_offset.clone())
                     .clobbers_flags(false)
-                    .compute_size("size_plus_maybe_sib_or_offset_for_in_reg_1")
+                    .compute_size("size_plus_maybe_sib_or_offset_for_inreg_1")
                     .emit(
                         r#"
                         if !flags.notrap() {
@@ -1383,7 +1486,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![fpr, gpr])
                 .inst_predicate(has_no_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_or_offset_for_in_reg_1")
+                .compute_size("size_plus_maybe_sib_or_offset_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1412,7 +1515,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![gpr, gpr])
                 .inst_predicate(has_small_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_1")
+                .compute_size("size_plus_maybe_sib_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1439,7 +1542,7 @@ pub(crate) fn define<'shared>(
                     .operands_in(vec![abcd, gpr])
                     .inst_predicate(has_small_offset.clone())
                     .clobbers_flags(false)
-                    .compute_size("size_plus_maybe_sib_for_in_reg_1")
+                    .compute_size("size_plus_maybe_sib_for_inreg_1")
                     .emit(
                         r#"
                         if !flags.notrap() {
@@ -1467,7 +1570,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![fpr, gpr])
                 .inst_predicate(has_small_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_1")
+                .compute_size("size_plus_maybe_sib_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1491,7 +1594,7 @@ pub(crate) fn define<'shared>(
             EncodingRecipeBuilder::new("stDisp32", &formats.store, 5)
                 .operands_in(vec![gpr, gpr])
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_1")
+                .compute_size("size_plus_maybe_sib_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1517,7 +1620,7 @@ pub(crate) fn define<'shared>(
                 EncodingRecipeBuilder::new("stDisp32_abcd", &formats.store, 5)
                     .operands_in(vec![abcd, gpr])
                     .clobbers_flags(false)
-                    .compute_size("size_plus_maybe_sib_for_in_reg_1")
+                    .compute_size("size_plus_maybe_sib_for_inreg_1")
                     .emit(
                         r#"
                         if !flags.notrap() {
@@ -1544,7 +1647,7 @@ pub(crate) fn define<'shared>(
             EncodingRecipeBuilder::new("fstDisp32", &formats.store, 5)
                 .operands_in(vec![fpr, gpr])
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_1")
+                .compute_size("size_plus_maybe_sib_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1577,7 +1680,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![gpr, gpr, gpr])
                 .inst_predicate(has_no_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_offset_for_in_reg_1")
+                .compute_size("size_plus_maybe_offset_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1604,7 +1707,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![abcd, gpr, gpr])
                 .inst_predicate(has_no_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_offset_for_in_reg_1")
+                .compute_size("size_plus_maybe_offset_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1630,7 +1733,7 @@ pub(crate) fn define<'shared>(
                 .operands_in(vec![fpr, gpr, gpr])
                 .inst_predicate(has_no_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_offset_for_in_reg_1")
+                .compute_size("size_plus_maybe_offset_for_inreg_1")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1867,7 +1970,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![gpr])
                 .inst_predicate(has_no_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_or_offset_for_in_reg_0")
+                .compute_size("size_plus_maybe_sib_or_offset_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1894,7 +1997,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![fpr])
                 .inst_predicate(has_no_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_or_offset_for_in_reg_0")
+                .compute_size("size_plus_maybe_sib_or_offset_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1924,7 +2027,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![gpr])
                 .inst_predicate(has_small_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_0")
+                .compute_size("size_plus_maybe_sib_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1950,7 +2053,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![fpr])
                 .inst_predicate(has_small_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_0")
+                .compute_size("size_plus_maybe_sib_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -1979,7 +2082,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![gpr])
                 .inst_predicate(has_big_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_0")
+                .compute_size("size_plus_maybe_sib_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -2005,7 +2108,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![fpr])
                 .inst_predicate(has_big_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_sib_for_in_reg_0")
+                .compute_size("size_plus_maybe_sib_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -2039,7 +2142,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![gpr])
                 .inst_predicate(has_no_offset.clone())
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_offset_for_in_reg_0")
+                .compute_size("size_plus_maybe_offset_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -2066,7 +2169,7 @@ pub(crate) fn define<'shared>(
                 .operands_out(vec![fpr])
                 .inst_predicate(has_no_offset)
                 .clobbers_flags(false)
-                .compute_size("size_plus_maybe_offset_for_in_reg_0")
+                .compute_size("size_plus_maybe_offset_for_inreg_0")
                 .emit(
                     r#"
                         if !flags.notrap() {
@@ -2396,7 +2499,7 @@ pub(crate) fn define<'shared>(
             .operands_out(vec![gpr])
             .clobbers_flags(false)
             .inst_predicate(valid_scale(&*formats.branch_table_entry))
-            .compute_size("size_plus_maybe_offset_for_in_reg_1")
+            .compute_size("size_plus_maybe_offset_for_inreg_1")
             .emit(
                 r#"
                     {{PUT_OP}}(bits, rex3(in_reg1, out_reg0, in_reg0), sink);
@@ -2475,7 +2578,7 @@ pub(crate) fn define<'shared>(
                 ),
             regs,
         )
-        .requires_prefix(true),
+        .rex_kind(RexRecipeKind::AlwaysEmitRex),
     );
 
     recipes.add_template(
@@ -2509,7 +2612,7 @@ pub(crate) fn define<'shared>(
                 ),
             regs,
         )
-        .requires_prefix(true),
+        .rex_kind(RexRecipeKind::AlwaysEmitRex),
     );
 
     recipes.add_template(
@@ -2532,7 +2635,8 @@ pub(crate) fn define<'shared>(
     // Conditional move (a.k.a integer select)
     // (maybe-REX.W) 0F 4x modrm(r,r)
     // 1 byte, modrm(r,r), is after the opcode
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("cmov", &formats.int_select, 1)
                 .operands_in(vec![
                     OperandConstraint::FixedReg(reg_rflags),
@@ -2547,10 +2651,14 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg1, in_reg2, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_cmov"),
     );
 
     // Bit scan forwards and reverse
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("bsf_and_bsr", &formats.unary, 1)
                 .operands_in(vec![gpr])
                 .operands_out(vec![
@@ -2563,12 +2671,16 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, out_reg0, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_outreg0"),
     );
 
     // Arithematic with flag I/O.
 
     // XX /r, MR form. Add two GPR registers and set carry flag.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("rout", &formats.binary, 1)
                 .operands_in(vec![gpr, gpr])
                 .operands_out(vec![
@@ -2582,10 +2694,14 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, in_reg1, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     // XX /r, MR form. Add two GPR registers and get carry flag.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("rin", &formats.ternary, 1)
                 .operands_in(vec![
                     OperandConstraint::RegClass(gpr),
@@ -2600,10 +2716,14 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, in_reg1, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     // XX /r, MR form. Add two GPR registers with carry flag.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("rio", &formats.ternary, 1)
                 .operands_in(vec![
                     OperandConstraint::RegClass(gpr),
@@ -2621,12 +2741,16 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, in_reg1, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     // Compare and set flags.
 
     // XX /r, MR form. Compare two GPR registers and set flags.
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("rcmp", &formats.binary, 1)
                 .operands_in(vec![gpr, gpr])
                 .operands_out(vec![reg_rflags])
@@ -2636,6 +2760,9 @@ pub(crate) fn define<'shared>(
                         modrm_rr(in_reg0, in_reg1, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     // Same as rcmp, but second operand is the stack pointer.
@@ -2669,7 +2796,8 @@ pub(crate) fn define<'shared>(
             InstructionPredicate::new_is_signed_int(&*formats.binary_imm, "imm", 8, 0);
 
         // XX /n, MI form with imm8.
-        recipes.add_template_recipe(
+        recipes.add_template(
+            Template::new(
                 EncodingRecipeBuilder::new("rcmp_ib", &formats.binary_imm, 2)
                     .operands_in(vec![gpr])
                     .operands_out(vec![reg_rflags])
@@ -2682,13 +2810,17 @@ pub(crate) fn define<'shared>(
                             sink.put1(imm as u8);
                         "#,
                     ),
+                regs,
+            )
+            .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
         );
 
         let has_big_offset =
             InstructionPredicate::new_is_signed_int(&*formats.binary_imm, "imm", 32, 0);
 
         // XX /n, MI form with imm32.
-        recipes.add_template_recipe(
+        recipes.add_template(
+            Template::new(
                 EncodingRecipeBuilder::new("rcmp_id", &formats.binary_imm, 5)
                     .operands_in(vec![gpr])
                     .operands_out(vec![reg_rflags])
@@ -2701,6 +2833,9 @@ pub(crate) fn define<'shared>(
                             sink.put4(imm as u32);
                         "#,
                     ),
+                regs,
+            )
+            .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
         );
     }
 
@@ -2718,7 +2853,8 @@ pub(crate) fn define<'shared>(
     // Bits 0-7 are the Jcc opcode.
     // Bits 8-15 control the test instruction which always has opcode byte 0x85.
 
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("tjccb", &formats.branch, 1 + 2)
                 .operands_in(vec![gpr])
                 .branch_range((3, 8))
@@ -2732,9 +2868,13 @@ pub(crate) fn define<'shared>(
                         disp1(destination, func, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
     );
 
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("tjccd", &formats.branch, 1 + 6)
                 .operands_in(vec![gpr])
                 .branch_range((7, 32))
@@ -2749,6 +2889,9 @@ pub(crate) fn define<'shared>(
                         disp4(destination, func, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
     );
 
     // 8-bit test-and-branch.
@@ -2770,7 +2913,7 @@ pub(crate) fn define<'shared>(
                 ),
             regs,
         )
-        .requires_prefix(true),
+        .rex_kind(RexRecipeKind::AlwaysEmitRex),
     );
 
     recipes.add_template(
@@ -2811,7 +2954,7 @@ pub(crate) fn define<'shared>(
                 ),
             regs,
         )
-        .requires_prefix(true),
+        .rex_kind(RexRecipeKind::AlwaysEmitRex),
     );
 
     recipes.add_template(
@@ -2878,7 +3021,8 @@ pub(crate) fn define<'shared>(
     // instruction, so it is limited to the `ABCD` register class for booleans.
     // The omission of a `when_prefixed` alternative is deliberate here.
 
-    recipes.add_template_recipe(
+    recipes.add_template(
+        Template::new(
             EncodingRecipeBuilder::new("icscc", &formats.int_compare, 1 + 3)
                 .operands_in(vec![gpr, gpr])
                 .operands_out(vec![abcd])
@@ -2894,6 +3038,9 @@ pub(crate) fn define<'shared>(
                         modrm_rr(out_reg0, 0, sink);
                     "#,
                 ),
+            regs,
+        )
+        .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0_inreg1"),
     );
 
     recipes.add_template_recipe(
@@ -2913,7 +3060,8 @@ pub(crate) fn define<'shared>(
         let is_small_imm =
             InstructionPredicate::new_is_signed_int(&*formats.int_compare_imm, "imm", 8, 0);
 
-        recipes.add_template_recipe(
+        recipes.add_template(
+            Template::new(
                 EncodingRecipeBuilder::new("icscc_ib", &formats.int_compare_imm, 2 + 3)
                     .operands_in(vec![gpr])
                     .operands_out(vec![abcd])
@@ -2932,12 +3080,16 @@ pub(crate) fn define<'shared>(
                             modrm_rr(out_reg0, 0, sink);
                         "#,
                     ),
+                regs,
+            )
+            .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
         );
 
         let is_big_imm =
             InstructionPredicate::new_is_signed_int(&*formats.int_compare_imm, "imm", 32, 0);
 
-        recipes.add_template_recipe(
+        recipes.add_template(
+            Template::new(
                 EncodingRecipeBuilder::new("icscc_id", &formats.int_compare_imm, 5 + 3)
                     .operands_in(vec![gpr])
                     .operands_out(vec![abcd])
@@ -2956,6 +3108,9 @@ pub(crate) fn define<'shared>(
                             modrm_rr(out_reg0, 0, sink);
                         "#,
                     ),
+                regs,
+            )
+            .inferred_rex_compute_size("size_with_inferred_rex_for_inreg0"),
         );
     }
 

cranelift/codegen/shared/src/isa/x86/encoding_bits.rs

@@ -57,6 +57,24 @@ impl EncodingBits {
         new
     }
 
+    /// Returns a copy of the EncodingBits with the RRR bits set.
+    #[inline]
+    pub fn with_rrr(self, rrr: u8) -> Self {
+        debug_assert_eq!(u8::from(self.rrr()), 0);
+        let mut enc = self.clone();
+        enc.write(RRR, rrr.into());
+        enc
+    }
+
+    /// Returns a copy of the EncodingBits with the REX.W bit set.
+    #[inline]
+    pub fn with_rex_w(self) -> Self {
+        debug_assert_eq!(self.rex_w(), 0);
+        let mut enc = self.clone();
+        enc.write(REX_W, 1);
+        enc
+    }
+
     /// Returns the raw bits.
     #[inline]
     pub fn bits(self) -> u16 {

cranelift/codegen/src/isa/x86/binemit.rs

@@ -61,6 +61,12 @@ fn rex3(rm: RegUnit, reg: RegUnit, index: RegUnit) -> u8 {
     BASE_REX | b | (x << 1) | (r << 2)
 }
 
+/// Determines whether a REX prefix should be emitted.
+#[inline]
+fn needs_rex(bits: u16, rex: u8) -> bool {
+    rex != BASE_REX || u8::from(EncodingBits::from(bits).rex_w()) == 1
+}
+
 // Emit a REX prefix.
 //
 // The R, X, and B bits are computed from registers using the functions above. The W bit is
@@ -80,11 +86,20 @@ fn put_op1<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
 
 // Emit a single-byte opcode with REX prefix.
 fn put_rexop1<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
-    debug_assert_eq!(bits & 0x0f00, 0, "Invalid encoding bits for Op1*");
+    debug_assert_eq!(bits & 0x0f00, 0, "Invalid encoding bits for RexOp1*");
     rex_prefix(bits, rex, sink);
     sink.put1(bits as u8);
 }
 
+/// Emit a single-byte opcode with inferred REX prefix.
+fn put_dynrexop1<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
+    debug_assert_eq!(bits & 0x0f00, 0, "Invalid encoding bits for DynRexOp1*");
+    if needs_rex(bits, rex) {
+        rex_prefix(bits, rex, sink);
+    }
+    sink.put1(bits as u8);
+}
+
 // Emit two-byte opcode: 0F XX
 fn put_op2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
     debug_assert_eq!(bits & 0x8f00, 0x0400, "Invalid encoding bits for Op2*");
@@ -101,6 +116,20 @@ fn put_rexop2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
     sink.put1(bits as u8);
 }
 
+/// Emit two-byte opcode: 0F XX with inferred REX prefix.
+fn put_dynrexop2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
+    debug_assert_eq!(
+        bits & 0x0f00,
+        0x0400,
+        "Invalid encoding bits for DynRexOp2*"
+    );
+    if needs_rex(bits, rex) {
+        rex_prefix(bits, rex, sink);
+    }
+    sink.put1(0x0f);
+    sink.put1(bits as u8);
+}
+
 // Emit single-byte opcode with mandatory prefix.
 fn put_mp1<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
     debug_assert_eq!(bits & 0x8c00, 0, "Invalid encoding bits for Mp1*");
@@ -112,7 +141,7 @@ fn put_mp1<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
 
 // Emit single-byte opcode with mandatory prefix and REX.
 fn put_rexmp1<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
-    debug_assert_eq!(bits & 0x0c00, 0, "Invalid encoding bits for Mp1*");
+    debug_assert_eq!(bits & 0x0c00, 0, "Invalid encoding bits for RexMp1*");
     let enc = EncodingBits::from(bits);
     sink.put1(PREFIX[(enc.pp() - 1) as usize]);
     rex_prefix(bits, rex, sink);
@@ -131,7 +160,7 @@ fn put_mp2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
 
 // Emit two-byte opcode (0F XX) with mandatory prefix and REX.
 fn put_rexmp2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
-    debug_assert_eq!(bits & 0x0c00, 0x0400, "Invalid encoding bits for Mp2*");
+    debug_assert_eq!(bits & 0x0c00, 0x0400, "Invalid encoding bits for RexMp2*");
     let enc = EncodingBits::from(bits);
     sink.put1(PREFIX[(enc.pp() - 1) as usize]);
     rex_prefix(bits, rex, sink);
@@ -139,6 +168,22 @@ fn put_rexmp2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
     sink.put1(bits as u8);
 }
 
+/// Emit two-byte opcode (0F XX) with mandatory prefix and inferred REX.
+fn put_dynrexmp2<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
+    debug_assert_eq!(
+        bits & 0x0c00,
+        0x0400,
+        "Invalid encoding bits for DynRexMp2*"
+    );
+    let enc = EncodingBits::from(bits);
+    sink.put1(PREFIX[(enc.pp() - 1) as usize]);
+    if needs_rex(bits, rex) {
+        rex_prefix(bits, rex, sink);
+    }
+    sink.put1(0x0f);
+    sink.put1(bits as u8);
+}
+
 // Emit three-byte opcode (0F 3[8A] XX) with mandatory prefix.
 fn put_mp3<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
     debug_assert_eq!(bits & 0x8800, 0x0800, "Invalid encoding bits for Mp3*");
@@ -152,7 +197,7 @@ fn put_mp3<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
 
 // Emit three-byte opcode (0F 3[8A] XX) with mandatory prefix and REX
 fn put_rexmp3<CS: CodeSink + ?Sized>(bits: u16, rex: u8, sink: &mut CS) {
-    debug_assert_eq!(bits & 0x0800, 0x0800, "Invalid encoding bits for Mp3*");
+    debug_assert_eq!(bits & 0x0800, 0x0800, "Invalid encoding bits for RexMp3*");
     let enc = EncodingBits::from(bits);
     sink.put1(PREFIX[(enc.pp() - 1) as usize]);
     rex_prefix(bits, rex, sink);

cranelift/codegen/src/isa/x86/enc_tables.rs

@@ -16,9 +16,20 @@ use crate::isa::{self, TargetIsa};
 use crate::predicates;
 use crate::regalloc::RegDiversions;
 
+use cranelift_codegen_shared::isa::x86::EncodingBits;
+
 include!(concat!(env!("OUT_DIR"), "/encoding-x86.rs"));
 include!(concat!(env!("OUT_DIR"), "/legalize-x86.rs"));
 
+/// Whether the REX prefix is needed for encoding extended registers (via REX.RXB).
+///
+/// Normal x86 instructions have only 3 bits for encoding a register.
+/// The REX prefix adds REX.R, REX,X, and REX.B bits, interpreted as fourth bits.
+pub fn is_extended_reg(reg: RegUnit) -> bool {
+    // Extended registers have the fourth bit set.
+    reg as u8 & 0b1000 != 0
+}
+
 pub fn needs_sib_byte(reg: RegUnit) -> bool {
     reg == RU::r12 as RegUnit || reg == RU::rsp as RegUnit
 }
@@ -29,74 +40,179 @@ pub fn needs_sib_byte_or_offset(reg: RegUnit) -> bool {
     needs_sib_byte(reg) || needs_offset(reg)
 }
 
-fn additional_size_if(
+fn test_input(
     op_index: usize,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
     condition_func: fn(RegUnit) -> bool,
-) -> u8 {
+) -> bool {
-    let addr_reg = divert.reg(func.dfg.inst_args(inst)[op_index], &func.locations);
+    let in_reg = divert.reg(func.dfg.inst_args(inst)[op_index], &func.locations);
-    if condition_func(addr_reg) {
+    condition_func(in_reg)
-        1
-    } else {
-        0
-    }
 }
 
-fn size_plus_maybe_offset_for_in_reg_0(
+fn test_result(
-    sizing: &RecipeSizing,
+    result_index: usize,
-    _enc: Encoding,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
-) -> u8 {
+    condition_func: fn(RegUnit) -> bool,
-    sizing.base_size + additional_size_if(0, inst, divert, func, needs_offset)
+) -> bool {
+    let out_reg = divert.reg(func.dfg.inst_results(inst)[result_index], &func.locations);
+    condition_func(out_reg)
 }
-fn size_plus_maybe_offset_for_in_reg_1(
+
+fn size_plus_maybe_offset_for_inreg_0(
     sizing: &RecipeSizing,
     _enc: Encoding,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
 ) -> u8 {
-    sizing.base_size + additional_size_if(1, inst, divert, func, needs_offset)
+    let needs_offset = test_input(0, inst, divert, func, needs_offset);
+    sizing.base_size + if needs_offset { 1 } else { 0 }
 }
-fn size_plus_maybe_sib_for_in_reg_0(
+fn size_plus_maybe_offset_for_inreg_1(
     sizing: &RecipeSizing,
     _enc: Encoding,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
 ) -> u8 {
-    sizing.base_size + additional_size_if(0, inst, divert, func, needs_sib_byte)
+    let needs_offset = test_input(1, inst, divert, func, needs_offset);
+    sizing.base_size + if needs_offset { 1 } else { 0 }
 }
-fn size_plus_maybe_sib_for_in_reg_1(
+fn size_plus_maybe_sib_for_inreg_0(
     sizing: &RecipeSizing,
     _enc: Encoding,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
 ) -> u8 {
-    sizing.base_size + additional_size_if(1, inst, divert, func, needs_sib_byte)
+    let needs_sib = test_input(0, inst, divert, func, needs_sib_byte);
+    sizing.base_size + if needs_sib { 1 } else { 0 }
 }
-fn size_plus_maybe_sib_or_offset_for_in_reg_0(
+fn size_plus_maybe_sib_for_inreg_1(
     sizing: &RecipeSizing,
     _enc: Encoding,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
 ) -> u8 {
-    sizing.base_size + additional_size_if(0, inst, divert, func, needs_sib_byte_or_offset)
+    let needs_sib = test_input(1, inst, divert, func, needs_sib_byte);
+    sizing.base_size + if needs_sib { 1 } else { 0 }
 }
-fn size_plus_maybe_sib_or_offset_for_in_reg_1(
+fn size_plus_maybe_sib_or_offset_for_inreg_0(
     sizing: &RecipeSizing,
     _enc: Encoding,
     inst: Inst,
     divert: &RegDiversions,
     func: &Function,
 ) -> u8 {
-    sizing.base_size + additional_size_if(1, inst, divert, func, needs_sib_byte_or_offset)
+    let needs_sib_or_offset = test_input(0, inst, divert, func, needs_sib_byte_or_offset);
+    sizing.base_size + if needs_sib_or_offset { 1 } else { 0 }
+}
+fn size_plus_maybe_sib_or_offset_for_inreg_1(
+    sizing: &RecipeSizing,
+    _enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_sib_or_offset = test_input(1, inst, divert, func, needs_sib_byte_or_offset);
+    sizing.base_size + if needs_sib_or_offset { 1 } else { 0 }
+}
+
+/// Infers whether a dynamic REX prefix will be emitted, for use with one input reg.
+///
+/// A REX prefix is known to be emitted if either:
+///  1. The EncodingBits specify that REX.W is to be set.
+///  2. Registers are used that require REX.R or REX.B bits for encoding.
+fn size_with_inferred_rex_for_inreg0(
+    sizing: &RecipeSizing,
+    enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_rex = (EncodingBits::from(enc.bits()).rex_w() != 0)
+        || test_input(0, inst, divert, func, is_extended_reg);
+    sizing.base_size + if needs_rex { 1 } else { 0 }
+}
+
+/// Infers whether a dynamic REX prefix will be emitted, based on the second operand.
+fn size_with_inferred_rex_for_inreg1(
+    sizing: &RecipeSizing,
+    enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_rex = (EncodingBits::from(enc.bits()).rex_w() != 0)
+        || test_input(1, inst, divert, func, is_extended_reg);
+    sizing.base_size + if needs_rex { 1 } else { 0 }
+}
+
+/// Infers whether a dynamic REX prefix will be emitted, based on the third operand.
+fn size_with_inferred_rex_for_inreg2(
+    sizing: &RecipeSizing,
+    enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_rex = (EncodingBits::from(enc.bits()).rex_w() != 0)
+        || test_input(2, inst, divert, func, is_extended_reg);
+    sizing.base_size + if needs_rex { 1 } else { 0 }
+}
+
+/// Infers whether a dynamic REX prefix will be emitted, for use with two input registers.
+///
+/// A REX prefix is known to be emitted if either:
+///  1. The EncodingBits specify that REX.W is to be set.
+///  2. Registers are used that require REX.R or REX.B bits for encoding.
+fn size_with_inferred_rex_for_inreg0_inreg1(
+    sizing: &RecipeSizing,
+    enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_rex = (EncodingBits::from(enc.bits()).rex_w() != 0)
+        || test_input(0, inst, divert, func, is_extended_reg)
+        || test_input(1, inst, divert, func, is_extended_reg);
+    sizing.base_size + if needs_rex { 1 } else { 0 }
+}
+
+/// Infers whether a dynamic REX prefix will be emitted, based on a single
+/// input register and a single output register.
+fn size_with_inferred_rex_for_inreg0_outreg0(
+    sizing: &RecipeSizing,
+    enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_rex = (EncodingBits::from(enc.bits()).rex_w() != 0)
+        || test_input(0, inst, divert, func, is_extended_reg)
+        || test_result(0, inst, divert, func, is_extended_reg);
+    sizing.base_size + if needs_rex { 1 } else { 0 }
+}
+
+/// Infers whether a dynamic REX prefix will be emitted, for use with CMOV.
+///
+/// CMOV uses 3 inputs, with the REX is inferred from reg1 and reg2.
+fn size_with_inferred_rex_for_cmov(
+    sizing: &RecipeSizing,
+    enc: Encoding,
+    inst: Inst,
+    divert: &RegDiversions,
+    func: &Function,
+) -> u8 {
+    let needs_rex = (EncodingBits::from(enc.bits()).rex_w() != 0)
+        || test_input(1, inst, divert, func, is_extended_reg)
+        || test_input(2, inst, divert, func, is_extended_reg);
+    sizing.base_size + if needs_rex { 1 } else { 0 }
 }
 
 /// If the value's definition is a constant immediate, returns its unpacked value, or None

cranelift/filetests/filetests/isa/x86/legalize-br-icmp.clif

@@ -15,7 +15,7 @@ ebb1:
 ; sameln: function %br_icmp(i64 [%rdi]) fast {
 ; nextln:                                 ebb0(v0: i64):
 ; nextln: [RexOp1pu_id#b8]                    v1 = iconst.i64 0
-; nextln: [RexOp1icscc#8039]                  v2 = icmp eq v0, v1
+; nextln: [DynRexOp1icscc#8039]               v2 = icmp eq v0, v1
 ; nextln: [RexOp1t8jccb#75]                   brnz v2, ebb1
 ; nextln: [Op1jmpb#eb]                        jump ebb1
 ; nextln: 
@@ -37,7 +37,7 @@ ebb1(v2: i64):
 ; sameln: function %br_icmp_ebb_args(i64 [%rdi]) fast {
 ; nextln:                                 ebb0(v0: i64):
 ; nextln: [RexOp1pu_id#b8]                    v1 = iconst.i64 0
-; nextln: [RexOp1icscc#8039]                  v3 = icmp eq v0, v1
+; nextln: [DynRexOp1icscc#8039]               v3 = icmp eq v0, v1
 ; nextln: [RexOp1t8jccb#75]                   brnz v3, ebb1(v0)
 ; nextln: [Op1jmpb#eb]                        jump ebb1(v0)
 ; nextln: 

cranelift/filetests/filetests/isa/x86/relax_branch.clif

@@ -22,94 +22,94 @@ function u0:2691(i32 [%rdi], i32 [%rsi], i64 vmctx [%r14]) -> i64 uext [%rax] ba
 
                                 ebb3(v8: i32 [%rdi], v19: i32 [%rsi]):
 @0005 [RexOp1ldDisp8#808b,%rax]        v7 = load.i64 v2+48
-@0005 [RexOp1rcmp_ib#f083,%rflags]  v91 = ifcmp_imm v7, 0
+@0005 [DynRexOp1rcmp_ib#f083,%rflags]  v91 = ifcmp_imm v7, 0
 @0005 [trapif#00]                      trapif ne v91, interrupt
-[Op1umr#89,%rax]                    v105 = copy v8
+[DynRexOp1umr#89,%rax]                 v105 = copy v8
-@000b [Op1r_ib#83,%rax]             v10 = iadd_imm v105, 1
+@000b [DynRexOp1r_ib#83,%rax]          v10 = iadd_imm v105, 1
                                        v80 -> v10
 @0010 [Op1umr#89,%rcx]                 v92 = uextend.i64 v8
 @0010 [RexOp1ld#808b,%rdx]             v93 = load.i64 notrap aligned readonly v2
                                        v95 -> v93
 @0010 [Op2ldWithIndex#4be,%rcx]        v12 = sload8_complex.i32 v93+v92
-[Op1umr#89,%rbx]                    v106 = copy v12
+[DynRexOp1umr#89,%rbx]                 v106 = copy v12
-@0017 [Op1r_ib#40c1,%rbx]           v14 = ishl_imm v106, 24
+@0017 [DynRexOp1r_ib#40c1,%rbx]        v14 = ishl_imm v106, 24
-@001a [Op1r_ib#70c1,%rbx]           v16 = sshr_imm v14, 24
+@001a [DynRexOp1r_ib#70c1,%rbx]        v16 = sshr_imm v14, 24
-[Op1umr#89,%rdi]                    v107 = copy v16
+[DynRexOp1umr#89,%rdi]                 v107 = copy v16
-@001f [Op1r_ib#83,%rdi]             v18 = iadd_imm v107, 32
+@001f [DynRexOp1r_ib#83,%rdi]          v18 = iadd_imm v107, 32
-[RexOp1umr#89,%r8]                  v108 = copy v19
+[DynRexOp1umr#89,%r8]                  v108 = copy v19
-@0026 [RexOp1r_ib#83,%r8]           v21 = iadd_imm v108, 1
+@0026 [DynRexOp1r_ib#83,%r8]           v21 = iadd_imm v108, 1
                                        v82 -> v21
 @002b [Op1umr#89,%rsi]                 v94 = uextend.i64 v19
 @002b [Op2ldWithIndex#4be,%rdx]        v23 = sload8_complex.i32 v93+v94
                                        v55 -> v23
-[Op1umr#89,%rsi]                    v109 = copy v23
+[DynRexOp1umr#89,%rsi]                 v109 = copy v23
-@0032 [Op1r_ib#40c1,%rsi]           v25 = ishl_imm v109, 24
+@0032 [DynRexOp1r_ib#40c1,%rsi]        v25 = ishl_imm v109, 24
-@0035 [Op1r_ib#70c1,%rsi]           v27 = sshr_imm v25, 24
+@0035 [DynRexOp1r_ib#70c1,%rsi]        v27 = sshr_imm v25, 24
                                        v69 -> v27
-[RexOp1umr#89,%r9]                  v110 = copy v27
+[DynRexOp1umr#89,%r9]                  v110 = copy v27
-@003a [RexOp1r_ib#83,%r9]           v29 = iadd_imm v110, 32
+@003a [DynRexOp1r_ib#83,%r9]           v29 = iadd_imm v110, 32
                                        v68 -> v29
-@0042 [Op1r_ib#83,%rcx]             v31 = iadd_imm v12, -65
+@0042 [DynRexOp1r_ib#83,%rcx]          v31 = iadd_imm v12, -65
-@0045 [Op1r_ib#40c1,%rcx]           v33 = ishl_imm v31, 24
+@0045 [DynRexOp1r_ib#40c1,%rcx]        v33 = ishl_imm v31, 24
-@0048 [Op1r_ib#70c1,%rcx]           v35 = sshr_imm v33, 24
+@0048 [DynRexOp1r_ib#70c1,%rcx]        v35 = sshr_imm v33, 24
-@004c [Op1r_id#4081,%rcx]           v37 = band_imm v35, 255
+@004c [DynRexOp1r_id#4081,%rcx]        v37 = band_imm v35, 255
-[Op1rcmp_ib#7083,%rflags]           v97 = ifcmp_imm v37, 26
+[DynRexOp1rcmp_ib#7083,%rflags]        v97 = ifcmp_imm v37, 26
 @0050 [Op1brib#70]                     brif sge v97, ebb6
 @0050 [-]                              fallthrough ebb10
 
                                 ebb10:
-[Op1umr#89,%rcx]                    v101 = copy v18
+[DynRexOp1umr#89,%rcx]              v101 = copy v18
 @0054 [Op1jmpb#eb]                  jump ebb5(v18, v101)
 
                                 ebb6:
-[Op1umr#89,%rcx]                    v102 = copy.i32 v16
+[DynRexOp1umr#89,%rcx]              v102 = copy.i32 v16
 @0059 [RexOp1rmov#89]               regmove v102, %rcx -> %rdi
 @0059 [RexOp1rmov#89]               regmove.i32 v16, %rbx -> %rcx
 @0059 [-]                           fallthrough ebb5(v102, v16)
 
                                 ebb5(v41: i32 [%rdi], v84: i32 [%rcx]):
                                     v83 -> v84
-@005d [Op1r_id#4081,%rdi]           v43 = band_imm v41, 255
+@005d [DynRexOp1r_id#4081,%rdi]     v43 = band_imm v41, 255
-@0062 [Op1r_ib#40c1,%rdi]           v45 = ishl_imm v43, 24
+@0062 [DynRexOp1r_ib#40c1,%rdi]     v45 = ishl_imm v43, 24
                                     v52 -> v45
 @0065 [RexOp1rmov#89]               regmove v45, %rdi -> %rbx
-@0065 [Op1r_ib#70c1,%rbx]           v47 = sshr_imm v45, 24
+@0065 [DynRexOp1r_ib#70c1,%rbx]     v47 = sshr_imm v45, 24
                                     v54 -> v47
 @0068 [RexOp1rmov#89]               regmove v47, %rbx -> %rdi
-@0068 [Op1icscc_ib#7083,%rbx]       v49 = icmp_imm ne v47, 0
+@0068 [DynRexOp1icscc_ib#7083,%rbx] v49 = icmp_imm ne v47, 0
 @0068 [RexOp2urm_noflags#4b6,%r10]  v50 = bint.i32 v49
-@0076 [Op1r_ib#83,%rdx]             v57 = iadd_imm.i32 v23, -65
+@0076 [DynRexOp1r_ib#83,%rdx]       v57 = iadd_imm.i32 v23, -65
-@0079 [Op1r_ib#40c1,%rdx]           v59 = ishl_imm v57, 24
+@0079 [DynRexOp1r_ib#40c1,%rdx]     v59 = ishl_imm v57, 24
-@007c [Op1r_ib#70c1,%rdx]           v61 = sshr_imm v59, 24
+@007c [DynRexOp1r_ib#70c1,%rdx]     v61 = sshr_imm v59, 24
-@0080 [Op1r_id#4081,%rdx]           v63 = band_imm v61, 255
+@0080 [DynRexOp1r_id#4081,%rdx]     v63 = band_imm v61, 255
-[Op1rcmp_ib#7083,%rflags]           v98 = ifcmp_imm v63, 26
+[DynRexOp1rcmp_ib#7083,%rflags]     v98 = ifcmp_imm v63, 26
 @0084 [RexOp1rmov#89]               regmove v47, %rdi -> %rbx
 @0084 [Op1brib#70]                  brif sge v98, ebb8
 @0084 [-]                           fallthrough ebb11
 
                                 ebb11:
-[RexOp1umr#89,%rdx]                 v103 = copy.i32 v29
+[DynRexOp1umr#89,%rdx]              v103 = copy.i32 v29
 @0088 [Op1jmpb#eb]                  jump ebb7(v29, v10, v21, v103)
 
                                 ebb8:
-[Op1umr#89,%rdx]                    v104 = copy.i32 v27
+[DynRexOp1umr#89,%rdx]              v104 = copy.i32 v27
 @008d [RexOp1rmov#89]               regmove v104, %rdx -> %r9
 @008d [RexOp1rmov#89]               regmove.i32 v27, %rsi -> %rdx
 @008d [-]                           fallthrough ebb7(v104, v10, v21, v27)
 
                                 ebb7(v67: i32 [%r9], v79: i32 [%rax], v81: i32 [%r8], v87: i32 [%rdx]):
-@0091 [RexOp1r_id#4081,%r9]         v71 = band_imm v67, 255
+@0091 [DynRexOp1r_id#4081,%r9]           v71 = band_imm v67, 255
-@0094 [RexOp1r_ib#40c1,%r9]         v73 = ishl_imm v71, 24
+@0094 [DynRexOp1r_ib#40c1,%r9]           v73 = ishl_imm v71, 24
-@0097 [RexOp1r_ib#70c1,%r9]         v75 = sshr_imm v73, 24
+@0097 [DynRexOp1r_ib#70c1,%r9]           v75 = sshr_imm v73, 24
-@0098 [RexOp1icscc#39,%rbx]         v76 = icmp.i32 eq v47, v75
+@0098 [DynRexOp1icscc#39,%rbx]           v76 = icmp.i32 eq v47, v75
 @0098 [Op2urm_noflags_abcd#4b6,%rbx]     v77 = bint.i32 v76
-@0099 [RexOp1rr#21,%r10]            v78 = band.i32 v50, v77
+@0099 [DynRexOp1rr#21,%r10]              v78 = band.i32 v50, v77
-@009a [RexOp1tjccb#74]              brz v78, ebb9
+@009a [DynRexOp1tjccb#74]                brz v78, ebb9
 @009a [-]                                fallthrough ebb12
 
                                 ebb12:
-[RexOp1umr#89,%rcx]                 v99 = copy v81
+[DynRexOp1umr#89,%rcx]              v99 = copy v81
-[Op1umr#89,%rdx]                    v100 = copy v79
+[DynRexOp1umr#89,%rdx]              v100 = copy v79
 @00a4 [RexOp1rmov#89]               regmove v100, %rdx -> %rdi
 @00a4 [RexOp1rmov#89]               regmove v99, %rcx -> %rsi
 @00a4 [Op1jmpd#e9]                  jump ebb3(v100, v99); bin: e9 ffffff2d
@@ -118,9 +118,9 @@ function u0:2691(i32 [%rdi], i32 [%rsi], i64 vmctx [%r14]) -> i64 uext [%rax] ba
 @00a7 [-]                           fallthrough ebb4
 
                                 ebb4:
-@00ad [Op1r_id#4081,%rcx]           v86 = band_imm.i32 v84, 255
+@00ad [DynRexOp1r_id#4081,%rcx]     v86 = band_imm.i32 v84, 255
-@00b3 [Op1r_id#4081,%rdx]           v89 = band_imm.i32 v87, 255
+@00b3 [DynRexOp1r_id#4081,%rdx]     v89 = band_imm.i32 v87, 255
-@00b4 [Op1rr#29,%rcx]               v90 = isub v86, v89
+@00b4 [DynRexOp1rr#29,%rcx]         v90 = isub v86, v89
 @00b5 [-]                           fallthrough ebb2(v90)
 
                                 ebb2(v5: i32 [%rcx]):

cranelift/filetests/filetests/isa/x86/shrink-multiple-uses.clif

@@ -4,7 +4,7 @@ target x86_64
 
 function %test_multiple_uses(i32 [%rdi]) -> i32 {
 ebb0(v0: i32 [%rdi]):
-[Op1rcmp_ib#7083,%rflags]           v3 = ifcmp_imm v0, 0
+[DynRexOp1rcmp_ib#7083,%rflags]     v3 = ifcmp_imm v0, 0
 [Op2seti_abcd#490,%rax]             v1 = trueif eq v3
 [RexOp2urm_noflags#4b6,%rax]        v2 = bint.i32 v1
 [Op1brib#70]                        brif eq v3, ebb1

cranelift/filetests/filetests/postopt/basic.clif

@@ -5,7 +5,7 @@ target i686
 
 function %br_icmp(i32, i32) -> i32 {
 ebb0(v0: i32, v1: i32):
-[Op1icscc#39,%rdx]  v2 = icmp slt v0, v1
+[DynRexOp1icscc#39,%rdx]  v2 = icmp slt v0, v1
 [Op1t8jccd_long#85]       brnz v2, ebb1
 [Op1jmpb#eb]              jump ebb2
 
@@ -35,7 +35,7 @@ ebb1:
 
 function %br_icmp_inverse(i32, i32) -> i32 {
 ebb0(v0: i32, v1: i32):
-[Op1icscc#39,%rdx]  v2 = icmp slt v0, v1
+[DynRexOp1icscc#39,%rdx]  v2 = icmp slt v0, v1
 [Op1t8jccd_long#84]       brz v2, ebb1
 [Op1jmpb#eb]              jump ebb2
 
@@ -65,7 +65,7 @@ ebb1:
 
 function %br_icmp_imm(i32, i32) -> i32 {
 ebb0(v0: i32, v1: i32):
-[Op1icscc_ib#7083]  v2 = icmp_imm slt v0, 2
+[DynRexOp1icscc_ib#7083]  v2 = icmp_imm slt v0, 2
 [Op1t8jccd_long#84]       brz v2, ebb1
 [Op1jmpb#eb]              jump ebb2
 

cranelift/filetests/filetests/postopt/complex_memory_ops.clif

@@ -3,7 +3,7 @@ target x86_64
 
 function %dual_loads(i64, i64) -> i64 {
 ebb0(v0: i64, v1: i64):
-[RexOp1rr#8001]    v3 = iadd v0, v1
+[DynRexOp1rr#8001] v3 = iadd v0, v1
                    v4 = load.i64 v3
                    v5 = uload8.i64 v3
                    v6 = sload8.i64 v3
@@ -29,7 +29,7 @@ ebb0(v0: i64, v1: i64):
 
 function %dual_loads2(i64, i64) -> i64 {
 ebb0(v0: i64, v1: i64):
-[RexOp1rr#8001]    v3 = iadd v0, v1
+[DynRexOp1rr#8001] v3 = iadd v0, v1
                    v4 = load.i64 v3+1
                    v5 = uload8.i64 v3+1
                    v6 = sload8.i64 v3+1
@@ -55,7 +55,7 @@ ebb0(v0: i64, v1: i64):
 
 function %dual_stores(i64, i64, i64) {
 ebb0(v0: i64, v1: i64, v2: i64):
-[RexOp1rr#8001]    v3 = iadd v0, v1
+[DynRexOp1rr#8001] v3 = iadd v0, v1
 [RexOp1st#8089]    store.i64 v2, v3
 [RexOp1st#88]      istore8.i64 v2, v3
 [RexMp1st#189]     istore16.i64 v2, v3
@@ -75,7 +75,7 @@ ebb0(v0: i64, v1: i64, v2: i64):
 
 function %dual_stores2(i64, i64, i64) {
 ebb0(v0: i64, v1: i64, v2: i64):
-[RexOp1rr#8001]         v3 = iadd v0, v1
+[DynRexOp1rr#8001] v3 = iadd v0, v1
 [RexOp1stDisp8#8089]    store.i64 v2, v3+1
 [RexOp1stDisp8#88]      istore8.i64 v2, v3+1
 [RexMp1stDisp8#189]     istore16.i64 v2, v3+1

cranelift/filetests/filetests/regalloc/coloring-227.clif

@@ -8,7 +8,7 @@ function %pr227(i32 [%rdi], i32 [%rsi], i32 [%rdx], i32 [%rcx], i64 vmctx [%r8])
                           ebb0(v0: i32, v1: i32, v2: i32, v3: i32, v4: i64):
 [RexOp1pu_id#b8]              v5 = iconst.i32 0
 [RexOp1pu_id#b8]              v6 = iconst.i32 0
-[RexOp1tjccb#74]              brz v6, ebb10
+[DynRexOp1tjccb#74]           brz v6, ebb10
 [Op1jmpb#eb]                  jump ebb3(v5, v5, v5, v5, v5, v5, v0, v1, v2, v3)
 
                           ebb3(v15: i32, v17: i32, v25: i32, v31: i32, v40: i32, v47: i32, v54: i32, v61: i32, v68: i32, v75: i32):
@@ -16,33 +16,33 @@ function %pr227(i32 [%rdi], i32 [%rsi], i32 [%rdx], i32 [%rcx], i64 vmctx [%r8])
 
                           ebb6:
 [RexOp1pu_id#b8]              v8 = iconst.i32 0
-[RexOp1tjccb#75]              brnz v8, ebb5
+[DynRexOp1tjccb#75]           brnz v8, ebb5
 [Op1jmpb#eb]                  jump ebb20
 
                           ebb20:
 [RexOp1pu_id#b8]              v9 = iconst.i32 0
 [RexOp1pu_id#b8]              v11 = iconst.i32 0
-[RexOp1icscc#39]              v12 = icmp.i32 eq v15, v11
+[DynRexOp1icscc#39]           v12 = icmp.i32 eq v15, v11
 [RexOp2urm_noflags#4b6]       v13 = bint.i32 v12
-[RexOp1rr#21]                 v14 = band v9, v13
+[DynRexOp1rr#21]              v14 = band v9, v13
-[RexOp1tjccb#75]              brnz v14, ebb6
+[DynRexOp1tjccb#75]           brnz v14, ebb6
 [Op1jmpb#eb]                  jump ebb7
 
                           ebb7:
-[RexOp1tjccb#74]              brz.i32 v17, ebb8
+[DynRexOp1tjccb#74]           brz.i32 v17, ebb8
 [Op1jmpb#eb]                  jump ebb17
 
                           ebb17:
 [RexOp1pu_id#b8]              v18 = iconst.i32 0
-[RexOp1tjccb#74]              brz v18, ebb9
+[DynRexOp1tjccb#74]           brz v18, ebb9
 [Op1jmpb#eb]                  jump ebb16
 
                           ebb16:
 [RexOp1pu_id#b8]              v21 = iconst.i32 0
 [RexOp1umr#89]                v79 = uextend.i64 v5
-[RexOp1r_ib#8083]             v80 = iadd_imm.i64 v4, 0
+[DynRexOp1r_ib#8083]          v80 = iadd_imm.i64 v4, 0
 [RexOp1ld#808b]               v81 = load.i64 v80
-[RexOp1rr#8001]               v22 = iadd v81, v79
+[DynRexOp1rr#8001]            v22 = iadd v81, v79
 [RexMp1st#189]                istore16 v21, v22
 [Op1jmpb#eb]                  jump ebb9
 
@@ -52,8 +52,8 @@ function %pr227(i32 [%rdi], i32 [%rsi], i32 [%rdx], i32 [%rcx], i64 vmctx [%r8])
                           ebb8:
 [RexOp1pu_id#b8]              v27 = iconst.i32 3
 [RexOp1pu_id#b8]              v28 = iconst.i32 4
-[RexOp1rr#09]                 v35 = bor.i32 v31, v13
+[DynRexOp1rr#09]              v35 = bor.i32 v31, v13
-[RexOp1tjccb#75]              brnz v35, ebb15(v27)
+[DynRexOp1tjccb#75]           brnz v35, ebb15(v27)
 [Op1jmpb#eb]                  jump ebb15(v28)
 
                           ebb15(v36: i32):
@@ -71,24 +71,24 @@ function %pr227(i32 [%rdi], i32 [%rsi], i32 [%rdx], i32 [%rcx], i64 vmctx [%r8])
 
                           ebb2(v7: i32, v45: i32, v52: i32, v59: i32, v66: i32, v73: i32):
 [RexOp1pu_id#b8]              v44 = iconst.i32 0
-[RexOp1tjccb#74]              brz v44, ebb12
+[DynRexOp1tjccb#74]           brz v44, ebb12
 [Op1jmpb#eb]                  jump ebb18
 
                           ebb18:
 [RexOp1pu_id#b8]              v50 = iconst.i32 11
-[RexOp1tjccb#74]              brz v50, ebb14
+[DynRexOp1tjccb#74]           brz v50, ebb14
 [Op1jmpb#eb]                  jump ebb19
 
                           ebb19:
 [RexOp1umr#89]                v82 = uextend.i64 v52
-[RexOp1r_ib#8083]             v83 = iadd_imm.i64 v4, 0
+[DynRexOp1r_ib#8083]          v83 = iadd_imm.i64 v4, 0
 [RexOp1ld#808b]               v84 = load.i64 v83
-[RexOp1rr#8001]               v57 = iadd v84, v82
+[DynRexOp1rr#8001]            v57 = iadd v84, v82
 [RexOp1ld#8b]                 v58 = load.i32 v57
 [RexOp1umr#89]                v85 = uextend.i64 v58
-[RexOp1r_ib#8083]             v86 = iadd_imm.i64 v4, 0
+[DynRexOp1r_ib#8083]          v86 = iadd_imm.i64 v4, 0
 [RexOp1ld#808b]               v87 = load.i64 v86
-[RexOp1rr#8001]               v64 = iadd v87, v85
+[DynRexOp1rr#8001]            v64 = iadd v87, v85
 [RexOp1st#88]                 istore8 v59, v64
 [RexOp1pu_id#b8]              v65 = iconst.i32 0
 [Op1jmpb#eb]                  jump ebb13(v65)
@@ -98,9 +98,9 @@ function %pr227(i32 [%rdi], i32 [%rsi], i32 [%rdx], i32 [%rcx], i64 vmctx [%r8])
 
                           ebb13(v51: i32):
 [RexOp1umr#89]                v88 = uextend.i64 v45
-[RexOp1r_ib#8083]             v89 = iadd_imm.i64 v4, 0
+[DynRexOp1r_ib#8083]          v89 = iadd_imm.i64 v4, 0
 [RexOp1ld#808b]               v90 = load.i64 v89
-[RexOp1rr#8001]               v71 = iadd v90, v88
+[DynRexOp1rr#8001]            v71 = iadd v90, v88
 [RexOp1st#89]                 store v51, v71
 [Op1jmpb#eb]                  jump ebb12
 

cranelift/filetests/filetests/verifier/flags.clif

@@ -4,7 +4,7 @@ target i686
 ; Simple, correct use of CPU flags.
 function %simple(i32) -> i32 {
                     ebb0(v0: i32):
-    [Op1rcmp#39]              v1 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]        v1 = ifcmp v0, v0
     [Op2seti_abcd#490]        v2 = trueif ugt v1
     [Op2urm_noflags_abcd#4b6] v3 = bint.i32 v2
     [Op1ret#c3]               return v3
@@ -13,7 +13,7 @@ function %simple(i32) -> i32 {
 ; Overlapping flag values of different types.
 function %overlap(i32, f32) -> i32 {
                     ebb0(v0: i32, v1: f32):
-    [Op1rcmp#39]              v2 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]        v2 = ifcmp v0, v0
     [Op2fcmp#42e]             v3 = ffcmp v1, v1
     [Op2setf_abcd#490]        v4 = trueff gt v3 ; error: conflicting live CPU flags: v2 and v3
     [Op2seti_abcd#490]        v5 = trueif ugt v2
@@ -25,8 +25,8 @@ function %overlap(i32, f32) -> i32 {
 ; CPU flags clobbered by arithmetic.
 function %clobbered(i32) -> i32 {
                     ebb0(v0: i32):
-    [Op1rcmp#39]              v1 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]        v1 = ifcmp v0, v0
-    [Op1rr#01]                v2 = iadd v0, v0 ; error: encoding clobbers live CPU flags in v1
+    [DynRexOp1rr#01]          v2 = iadd v0, v0 ; error: encoding clobbers live CPU flags in v1
     [Op2seti_abcd#490]        v3 = trueif ugt v1
     [Op2urm_noflags_abcd#4b6] v4 = bint.i32 v3
     [Op1ret#c3]               return v4
@@ -35,7 +35,7 @@ function %clobbered(i32) -> i32 {
 ; CPU flags not clobbered by load.
 function %live_across_load(i32) -> i32 {
                     ebb0(v0: i32):
-    [Op1rcmp#39]              v1 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]        v1 = ifcmp v0, v0
     [Op1ld#8b]                v2 = load.i32 v0
     [Op2seti_abcd#490]        v3 = trueif ugt v1
     [Op2urm_noflags_abcd#4b6] v4 = bint.i32 v3
@@ -45,7 +45,7 @@ function %live_across_load(i32) -> i32 {
 ; Correct use of CPU flags across EBB.
 function %live_across_ebb(i32) -> i32 {
                           ebb0(v0: i32):
-    [Op1rcmp#39]              v1 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]        v1 = ifcmp v0, v0
     [Op1jmpb#eb]              jump ebb1
                           ebb1:
     [Op2seti_abcd#490]        v2 = trueif ugt v1
@@ -61,14 +61,14 @@ function %live_across_ebb_backwards(i32) -> i32 {
     [Op2urm_noflags_abcd#4b6] v3 = bint.i32 v2
     [Op1ret#c3]               return v3
                           ebb2:
-    [Op1rcmp#39]              v1 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]        v1 = ifcmp v0, v0
     [Op1jmpb#eb]              jump ebb1
 }
 
 ; Flags live into loop.
 function %live_into_loop(i32) -> i32 {
                     ebb0(v0: i32):
-    [Op1rcmp#39]        v1 = ifcmp v0, v0
+    [DynRexOp1rcmp#39]  v1 = ifcmp v0, v0
     [Op1jmpb#eb]        jump ebb1
                     ebb1:
     [Op2seti_abcd#490]  v2 = trueif ugt v1