diff --git a/cranelift/codegen/meta/src/isa/riscv/encodings.rs b/cranelift/codegen/meta/src/isa/riscv/encodings.rs
index a5af1214aa..af5e471a53 100644
--- a/cranelift/codegen/meta/src/isa/riscv/encodings.rs
+++ b/cranelift/codegen/meta/src/isa/riscv/encodings.rs
@@ -230,8 +230,7 @@ pub fn define<'defs>(
          -> InstructionPredicateNode {
             let x = var_pool.create("x");
             let y = var_pool.create("y");
-            let cc =
-                Literal::enumerator_for(shared_defs.operand_kinds.by_name("intcc"), intcc_field);
+            let cc = Literal::enumerator_for(&shared_defs.imm.intcc, intcc_field);
             Apply::new(
                 bound_inst.clone().into(),
                 vec![Expr::Literal(cc), Expr::Var(x), Expr::Var(y)],
@@ -313,8 +312,7 @@ pub fn define<'defs>(
             let y = var_pool.create("y");
             let dest = var_pool.create("dest");
             let args = var_pool.create("args");
-            let cc =
-                Literal::enumerator_for(shared_defs.operand_kinds.by_name("intcc"), intcc_field);
+            let cc = Literal::enumerator_for(&shared_defs.imm.intcc, intcc_field);
             Apply::new(
                 bound_inst.clone().into(),
                 vec![
diff --git a/cranelift/codegen/meta/src/isa/x86/instructions.rs b/cranelift/codegen/meta/src/isa/x86/instructions.rs
index 6464261938..742b632642 100644
--- a/cranelift/codegen/meta/src/isa/x86/instructions.rs
+++ b/cranelift/codegen/meta/src/isa/x86/instructions.rs
@@ -7,11 +7,13 @@ use crate::cdsl::instructions::{
 use crate::cdsl::operands::{create_operand as operand, create_operand_doc as operand_doc};
 use crate::cdsl::types::ValueType;
 use crate::cdsl::typevar::{Interval, TypeSetBuilder, TypeVar};
-use crate::shared::{immediates, types, OperandKinds};
+use crate::shared::immediates::Immediates;
+use crate::shared::types;
 
 pub fn define(
     mut all_instructions: &mut AllInstructions,
     format_registry: &FormatRegistry,
+    immediates: &Immediates,
 ) -> InstructionGroup {
     let mut ig = InstructionGroupBuilder::new(
         "x86",
@@ -249,8 +251,7 @@ pub fn define(
         .operands_out(vec![y, rflags]),
     );
 
-    let immediates = OperandKinds::from(immediates::define());
-    let uimm8 = immediates.by_name("uimm8");
+    let uimm8 = &immediates.uimm8;
     let TxN = &TypeVar::new(
         "TxN",
         "A SIMD vector type",
diff --git a/cranelift/codegen/meta/src/isa/x86/legalize.rs b/cranelift/codegen/meta/src/isa/x86/legalize.rs
index a64f631168..158b05e4f1 100644
--- a/cranelift/codegen/meta/src/isa/x86/legalize.rs
+++ b/cranelift/codegen/meta/src/isa/x86/legalize.rs
@@ -58,12 +58,7 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
     let x86_umulx = x86_instructions.by_name("x86_umulx");
     let x86_smulx = x86_instructions.by_name("x86_smulx");
 
-    // List of immediates.
-    let floatcc = shared.operand_kinds.by_name("floatcc");
-    let imm64 = shared.operand_kinds.by_name("imm64");
-    let intcc = shared.operand_kinds.by_name("intcc");
-    let uimm8 = shared.operand_kinds.by_name("uimm8");
-    let ieee64 = shared.operand_kinds.by_name("ieee64");
+    let imm = &shared.imm;
 
     // Division and remainder.
     //
@@ -99,12 +94,12 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
     // `ucomiss` or `ucomisd` instruction. The remaining codes need legalization
     // patterns.
 
-    let floatcc_eq = Literal::enumerator_for(floatcc, "eq");
-    let floatcc_ord = Literal::enumerator_for(floatcc, "ord");
-    let floatcc_ueq = Literal::enumerator_for(floatcc, "ueq");
-    let floatcc_ne = Literal::enumerator_for(floatcc, "ne");
-    let floatcc_uno = Literal::enumerator_for(floatcc, "uno");
-    let floatcc_one = Literal::enumerator_for(floatcc, "one");
+    let floatcc_eq = Literal::enumerator_for(&imm.floatcc, "eq");
+    let floatcc_ord = Literal::enumerator_for(&imm.floatcc, "ord");
+    let floatcc_ueq = Literal::enumerator_for(&imm.floatcc, "ueq");
+    let floatcc_ne = Literal::enumerator_for(&imm.floatcc, "ne");
+    let floatcc_uno = Literal::enumerator_for(&imm.floatcc, "uno");
+    let floatcc_one = Literal::enumerator_for(&imm.floatcc, "one");
 
     // Equality needs an explicit `ord` test which checks the parity bit.
     group.legalize(
@@ -124,14 +119,14 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
         ],
     );
 
-    let floatcc_lt = &Literal::enumerator_for(floatcc, "lt");
-    let floatcc_gt = &Literal::enumerator_for(floatcc, "gt");
-    let floatcc_le = &Literal::enumerator_for(floatcc, "le");
-    let floatcc_ge = &Literal::enumerator_for(floatcc, "ge");
-    let floatcc_ugt = &Literal::enumerator_for(floatcc, "ugt");
-    let floatcc_ult = &Literal::enumerator_for(floatcc, "ult");
-    let floatcc_uge = &Literal::enumerator_for(floatcc, "uge");
-    let floatcc_ule = &Literal::enumerator_for(floatcc, "ule");
+    let floatcc_lt = &Literal::enumerator_for(&imm.floatcc, "lt");
+    let floatcc_gt = &Literal::enumerator_for(&imm.floatcc, "gt");
+    let floatcc_le = &Literal::enumerator_for(&imm.floatcc, "le");
+    let floatcc_ge = &Literal::enumerator_for(&imm.floatcc, "ge");
+    let floatcc_ugt = &Literal::enumerator_for(&imm.floatcc, "ugt");
+    let floatcc_ult = &Literal::enumerator_for(&imm.floatcc, "ult");
+    let floatcc_uge = &Literal::enumerator_for(&imm.floatcc, "uge");
+    let floatcc_ule = &Literal::enumerator_for(&imm.floatcc, "ule");
 
     // Inequalities that need to be reversed.
     for &(cc, rev_cc) in &[
@@ -165,9 +160,9 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
     let r2flags = var("r2flags");
     let index2 = var("index2");
 
-    let intcc_eq = Literal::enumerator_for(intcc, "eq");
-    let imm64_minus_one = Literal::constant(imm64, -1);
-    let imm64_63 = Literal::constant(imm64, 63);
+    let intcc_eq = Literal::enumerator_for(&imm.intcc, "eq");
+    let imm64_minus_one = Literal::constant(&imm.imm64, -1);
+    let imm64_63 = Literal::constant(&imm.imm64, 63);
     group.legalize(
         def!(a = clz.I64(x)),
         vec![
@@ -179,7 +174,7 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
         ],
     );
 
-    let imm64_31 = Literal::constant(imm64, 31);
+    let imm64_31 = Literal::constant(&imm.imm64, 31);
     group.legalize(
         def!(a = clz.I32(x)),
         vec![
@@ -191,7 +186,7 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
         ],
     );
 
-    let imm64_64 = Literal::constant(imm64, 64);
+    let imm64_64 = Literal::constant(&imm.imm64, 64);
     group.legalize(
         def!(a = ctz.I64(x)),
         vec![
@@ -201,7 +196,7 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
         ],
     );
 
-    let imm64_32 = Literal::constant(imm64, 32);
+    let imm64_32 = Literal::constant(&imm.imm64, 32);
     group.legalize(
         def!(a = ctz.I32(x)),
         vec![
@@ -232,13 +227,13 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
     let qc0F = var("qc0F");
     let qc01 = var("qc01");
 
-    let imm64_1 = Literal::constant(imm64, 1);
-    let imm64_4 = Literal::constant(imm64, 4);
+    let imm64_1 = Literal::constant(&imm.imm64, 1);
+    let imm64_4 = Literal::constant(&imm.imm64, 4);
     group.legalize(
         def!(qv16 = popcnt.I64(qv1)),
         vec![
             def!(qv3 = ushr_imm(qv1, imm64_1)),
-            def!(qc77 = iconst(Literal::constant(imm64, 0x7777777777777777))),
+            def!(qc77 = iconst(Literal::constant(&imm.imm64, 0x7777777777777777))),
             def!(qv4 = band(qv3, qc77)),
             def!(qv5 = isub(qv1, qv4)),
             def!(qv6 = ushr_imm(qv4, imm64_1)),
@@ -249,11 +244,11 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
             def!(qv11 = isub(qv8, qv10)),
             def!(qv12 = ushr_imm(qv11, imm64_4)),
             def!(qv13 = iadd(qv11, qv12)),
-            def!(qc0F = iconst(Literal::constant(imm64, 0x0F0F0F0F0F0F0F0F))),
+            def!(qc0F = iconst(Literal::constant(&imm.imm64, 0x0F0F0F0F0F0F0F0F))),
             def!(qv14 = band(qv13, qc0F)),
-            def!(qc01 = iconst(Literal::constant(imm64, 0x0101010101010101))),
+            def!(qc01 = iconst(Literal::constant(&imm.imm64, 0x0101010101010101))),
             def!(qv15 = imul(qv14, qc01)),
-            def!(qv16 = ushr_imm(qv15, Literal::constant(imm64, 56))),
+            def!(qv16 = ushr_imm(qv15, Literal::constant(&imm.imm64, 56))),
         ],
     );
 
@@ -281,7 +276,7 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
         def!(lv16 = popcnt.I32(lv1)),
         vec![
             def!(lv3 = ushr_imm(lv1, imm64_1)),
-            def!(lc77 = iconst(Literal::constant(imm64, 0x77777777))),
+            def!(lc77 = iconst(Literal::constant(&imm.imm64, 0x77777777))),
             def!(lv4 = band(lv3, lc77)),
             def!(lv5 = isub(lv1, lv4)),
             def!(lv6 = ushr_imm(lv4, imm64_1)),
@@ -292,11 +287,11 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
             def!(lv11 = isub(lv8, lv10)),
             def!(lv12 = ushr_imm(lv11, imm64_4)),
             def!(lv13 = iadd(lv11, lv12)),
-            def!(lc0F = iconst(Literal::constant(imm64, 0x0F0F0F0F))),
+            def!(lc0F = iconst(Literal::constant(&imm.imm64, 0x0F0F0F0F))),
             def!(lv14 = band(lv13, lc0F)),
-            def!(lc01 = iconst(Literal::constant(imm64, 0x01010101))),
+            def!(lc01 = iconst(Literal::constant(&imm.imm64, 0x01010101))),
             def!(lv15 = imul(lv14, lc01)),
-            def!(lv16 = ushr_imm(lv15, Literal::constant(imm64, 24))),
+            def!(lv16 = ushr_imm(lv15, Literal::constant(&imm.imm64, 24))),
         ],
     );
 
@@ -313,9 +308,9 @@ pub fn define(shared: &mut SharedDefinitions, x86_instructions: &InstructionGrou
     .chain_with(shared.transform_groups.by_name("narrow").id);
 
     // SIMD
-    let uimm8_zero = Literal::constant(uimm8, 0x00);
-    let uimm8_one = Literal::constant(uimm8, 0x01);
-    let ieee64_zero = Literal::constant(ieee64, 0x00);
+    let uimm8_zero = Literal::constant(&imm.uimm8, 0x00);
+    let uimm8_one = Literal::constant(&imm.uimm8, 0x01);
+    let ieee64_zero = Literal::constant(&imm.ieee64, 0x00);
     let b = var("b");
     let c = var("c");
     let d = var("d");
diff --git a/cranelift/codegen/meta/src/isa/x86/mod.rs b/cranelift/codegen/meta/src/isa/x86/mod.rs
index 9f4a4a334a..5122c7b2ad 100644
--- a/cranelift/codegen/meta/src/isa/x86/mod.rs
+++ b/cranelift/codegen/meta/src/isa/x86/mod.rs
@@ -20,6 +20,7 @@ pub fn define(shared_defs: &mut SharedDefinitions) -> TargetIsa {
     let inst_group = instructions::define(
         &mut shared_defs.all_instructions,
         &shared_defs.format_registry,
+        &shared_defs.imm,
     );
     legalize::define(shared_defs, &inst_group);
 
diff --git a/cranelift/codegen/meta/src/isa/x86/recipes.rs b/cranelift/codegen/meta/src/isa/x86/recipes.rs
index 579da62bdd..8eb0fdac6d 100644
--- a/cranelift/codegen/meta/src/isa/x86/recipes.rs
+++ b/cranelift/codegen/meta/src/isa/x86/recipes.rs
@@ -332,7 +332,7 @@ pub fn define<'shared>(
 ) -> RecipeGroup<'shared> {
     // The set of floating point condition codes that are directly supported.
     // Other condition codes need to be reversed or expressed as two tests.
-    let floatcc = shared_defs.operand_kinds.by_name("floatcc");
+    let floatcc = &shared_defs.imm.floatcc;
     let supported_floatccs: Vec<Literal> = ["ord", "uno", "one", "ueq", "gt", "ge", "ult", "ule"]
         .iter()
         .map(|name| Literal::enumerator_for(floatcc, name))
diff --git a/cranelift/codegen/meta/src/shared/formats.rs b/cranelift/codegen/meta/src/shared/formats.rs
index 2a16f250a1..2f79fdbdf1 100644
--- a/cranelift/codegen/meta/src/shared/formats.rs
+++ b/cranelift/codegen/meta/src/shared/formats.rs
@@ -1,22 +1,8 @@
 use crate::cdsl::formats::{FormatRegistry, InstructionFormatBuilder as Builder};
+use crate::shared::immediates::Immediates;
 use crate::shared::OperandKinds;
 
-pub fn define(immediates: &OperandKinds, entities: &OperandKinds) -> FormatRegistry {
-    // Shorthands for immediates.
-    let uimm8 = immediates.by_name("uimm8");
-    let uimm32 = immediates.by_name("uimm32");
-    let uimm128 = immediates.by_name("uimm128");
-    let imm64 = immediates.by_name("imm64");
-    let ieee32 = immediates.by_name("ieee32");
-    let ieee64 = immediates.by_name("ieee64");
-    let boolean = immediates.by_name("boolean");
-    let intcc = immediates.by_name("intcc");
-    let floatcc = immediates.by_name("floatcc");
-    let memflags = immediates.by_name("memflags");
-    let offset32 = immediates.by_name("offset32");
-    let trapcode = immediates.by_name("trapcode");
-    let regunit = immediates.by_name("regunit");
-
+pub fn define(imm: &Immediates, entities: &OperandKinds) -> FormatRegistry {
     // Shorthands for entities.
     let global_value = entities.by_name("global_value");
     let ebb = entities.by_name("ebb");
@@ -30,15 +16,15 @@ pub fn define(immediates: &OperandKinds, entities: &OperandKinds) -> FormatRegis
     let mut registry = FormatRegistry::new();
 
     registry.insert(Builder::new("Unary").value());
-    registry.insert(Builder::new("UnaryImm").imm(imm64));
-    registry.insert(Builder::new("UnaryImm128").imm(uimm128));
-    registry.insert(Builder::new("UnaryIeee32").imm(ieee32));
-    registry.insert(Builder::new("UnaryIeee64").imm(ieee64));
-    registry.insert(Builder::new("UnaryBool").imm(boolean));
+    registry.insert(Builder::new("UnaryImm").imm(&imm.imm64));
+    registry.insert(Builder::new("UnaryImm128").imm(&imm.uimm128));
+    registry.insert(Builder::new("UnaryIeee32").imm(&imm.ieee32));
+    registry.insert(Builder::new("UnaryIeee64").imm(&imm.ieee64));
+    registry.insert(Builder::new("UnaryBool").imm(&imm.boolean));
     registry.insert(Builder::new("UnaryGlobalValue").imm(global_value));
 
     registry.insert(Builder::new("Binary").value().value());
-    registry.insert(Builder::new("BinaryImm").value().imm(imm64));
+    registry.insert(Builder::new("BinaryImm").value().imm(&imm.imm64));
 
     // The select instructions are controlled by the second VALUE operand.
     // The first VALUE operand is the controlling flag which has a derived type.
@@ -60,43 +46,59 @@ pub fn define(immediates: &OperandKinds, entities: &OperandKinds) -> FormatRegis
     registry.insert(
         Builder::new("InsertLane")
             .value()
-            .imm_with_name("lane", uimm8)
+            .imm_with_name("lane", &imm.uimm8)
             .value(),
     );
     registry.insert(
         Builder::new("ExtractLane")
             .value()
-            .imm_with_name("lane", uimm8),
+            .imm_with_name("lane", &imm.uimm8),
     );
 
-    registry.insert(Builder::new("IntCompare").imm(intcc).value().value());
-    registry.insert(Builder::new("IntCompareImm").imm(intcc).value().imm(imm64));
-    registry.insert(Builder::new("IntCond").imm(intcc).value());
+    registry.insert(Builder::new("IntCompare").imm(&imm.intcc).value().value());
+    registry.insert(
+        Builder::new("IntCompareImm")
+            .imm(&imm.intcc)
+            .value()
+            .imm(&imm.imm64),
+    );
+    registry.insert(Builder::new("IntCond").imm(&imm.intcc).value());
 
-    registry.insert(Builder::new("FloatCompare").imm(floatcc).value().value());
-    registry.insert(Builder::new("FloatCond").imm(floatcc).value());;
+    registry.insert(
+        Builder::new("FloatCompare")
+            .imm(&imm.floatcc)
+            .value()
+            .value(),
+    );
+    registry.insert(Builder::new("FloatCond").imm(&imm.floatcc).value());;
 
-    registry.insert(Builder::new("IntSelect").imm(intcc).value().value().value());
+    registry.insert(
+        Builder::new("IntSelect")
+            .imm(&imm.intcc)
+            .value()
+            .value()
+            .value(),
+    );
 
     registry.insert(Builder::new("Jump").imm(ebb).varargs());
     registry.insert(Builder::new("Branch").value().imm(ebb).varargs());
     registry.insert(
         Builder::new("BranchInt")
-            .imm(intcc)
+            .imm(&imm.intcc)
             .value()
             .imm(ebb)
             .varargs(),
     );
     registry.insert(
         Builder::new("BranchFloat")
-            .imm(floatcc)
+            .imm(&imm.floatcc)
             .value()
             .imm(ebb)
             .varargs(),
     );
     registry.insert(
         Builder::new("BranchIcmp")
-            .imm(intcc)
+            .imm(&imm.intcc)
             .value()
             .value()
             .imm(ebb)
@@ -107,7 +109,7 @@ pub fn define(immediates: &OperandKinds, entities: &OperandKinds) -> FormatRegis
         Builder::new("BranchTableEntry")
             .value()
             .value()
-            .imm(uimm8)
+            .imm(&imm.uimm8)
             .imm(jump_table),
     );
     registry.insert(Builder::new("BranchTableBase").imm(jump_table));
@@ -117,74 +119,89 @@ pub fn define(immediates: &OperandKinds, entities: &OperandKinds) -> FormatRegis
     registry.insert(Builder::new("CallIndirect").imm(sig_ref).value().varargs());
     registry.insert(Builder::new("FuncAddr").imm(func_ref));
 
-    registry.insert(Builder::new("Load").imm(memflags).value().imm(offset32));
+    registry.insert(
+        Builder::new("Load")
+            .imm(&imm.memflags)
+            .value()
+            .imm(&imm.offset32),
+    );
     registry.insert(
         Builder::new("LoadComplex")
-            .imm(memflags)
+            .imm(&imm.memflags)
             .varargs()
-            .imm(offset32),
+            .imm(&imm.offset32),
     );
     registry.insert(
         Builder::new("Store")
-            .imm(memflags)
+            .imm(&imm.memflags)
             .value()
             .value()
-            .imm(offset32),
+            .imm(&imm.offset32),
     );
     registry.insert(
         Builder::new("StoreComplex")
-            .imm(memflags)
+            .imm(&imm.memflags)
             .value()
             .varargs()
-            .imm(offset32),
+            .imm(&imm.offset32),
     );
-    registry.insert(Builder::new("StackLoad").imm(stack_slot).imm(offset32));
+    registry.insert(Builder::new("StackLoad").imm(stack_slot).imm(&imm.offset32));
     registry.insert(
         Builder::new("StackStore")
             .value()
             .imm(stack_slot)
-            .imm(offset32),
+            .imm(&imm.offset32),
     );
 
     // Accessing a WebAssembly heap.
-    registry.insert(Builder::new("HeapAddr").imm(heap).value().imm(uimm32));
+    registry.insert(Builder::new("HeapAddr").imm(heap).value().imm(&imm.uimm32));
 
     // Accessing a WebAssembly table.
-    registry.insert(Builder::new("TableAddr").imm(table).value().imm(offset32));
+    registry.insert(
+        Builder::new("TableAddr")
+            .imm(table)
+            .value()
+            .imm(&imm.offset32),
+    );
 
     registry.insert(
         Builder::new("RegMove")
             .value()
-            .imm_with_name("src", regunit)
-            .imm_with_name("dst", regunit),
+            .imm_with_name("src", &imm.regunit)
+            .imm_with_name("dst", &imm.regunit),
     );
     registry.insert(
         Builder::new("CopySpecial")
-            .imm_with_name("src", regunit)
-            .imm_with_name("dst", regunit),
+            .imm_with_name("src", &imm.regunit)
+            .imm_with_name("dst", &imm.regunit),
     );
-    registry.insert(Builder::new("CopyToSsa").imm_with_name("src", regunit));
+    registry.insert(Builder::new("CopyToSsa").imm_with_name("src", &imm.regunit));
     registry.insert(
         Builder::new("RegSpill")
             .value()
-            .imm_with_name("src", regunit)
+            .imm_with_name("src", &imm.regunit)
             .imm_with_name("dst", stack_slot),
     );
     registry.insert(
         Builder::new("RegFill")
             .value()
             .imm_with_name("src", stack_slot)
-            .imm_with_name("dst", regunit),
+            .imm_with_name("dst", &imm.regunit),
     );
 
-    registry.insert(Builder::new("Trap").imm(trapcode));
-    registry.insert(Builder::new("CondTrap").value().imm(trapcode));
-    registry.insert(Builder::new("IntCondTrap").imm(intcc).value().imm(trapcode));
+    registry.insert(Builder::new("Trap").imm(&imm.trapcode));
+    registry.insert(Builder::new("CondTrap").value().imm(&imm.trapcode));
+    registry.insert(
+        Builder::new("IntCondTrap")
+            .imm(&imm.intcc)
+            .value()
+            .imm(&imm.trapcode),
+    );
     registry.insert(
         Builder::new("FloatCondTrap")
-            .imm(floatcc)
+            .imm(&imm.floatcc)
             .value()
-            .imm(trapcode),
+            .imm(&imm.trapcode),
     );
 
     registry
diff --git a/cranelift/codegen/meta/src/shared/immediates.rs b/cranelift/codegen/meta/src/shared/immediates.rs
index 5b8baca898..b7a7f712a2 100644
--- a/cranelift/codegen/meta/src/shared/immediates.rs
+++ b/cranelift/codegen/meta/src/shared/immediates.rs
@@ -2,154 +2,174 @@ use crate::cdsl::operands::{OperandKind, OperandKindBuilder as Builder};
 
 use std::collections::HashMap;
 
-pub fn define() -> Vec<OperandKind> {
-    let mut kinds = Vec::new();
+pub struct Immediates {
+    /// A 64-bit immediate integer operand.
+    ///
+    /// This type of immediate integer can interact with SSA values with any IntType type.
+    pub imm64: OperandKind,
 
-    // A 64-bit immediate integer operand.
-    //
-    // This type of immediate integer can interact with SSA values with any
-    // IntType type.
-    let imm64 = Builder::new_imm("imm64")
-        .doc("A 64-bit immediate integer.")
-        .build();
-    kinds.push(imm64);
+    /// An unsigned 8-bit immediate integer operand.
+    ///
+    /// This small operand is used to indicate lane indexes in SIMD vectors and immediate bit
+    /// counts on shift instructions.
+    pub uimm8: OperandKind,
 
-    // An unsigned 8-bit immediate integer operand.
-    //
-    // This small operand is used to indicate lane indexes in SIMD vectors and
-    // immediate bit counts on shift instructions.
-    let uimm8 = Builder::new_imm("uimm8")
-        .doc("An 8-bit immediate unsigned integer.")
-        .build();
-    kinds.push(uimm8);
+    /// An unsigned 32-bit immediate integer operand.
+    pub uimm32: OperandKind,
 
-    // An unsigned 32-bit immediate integer operand.
-    let uimm32 = Builder::new_imm("uimm32")
-        .doc("A 32-bit immediate unsigned integer.")
-        .build();
-    kinds.push(uimm32);
+    /// An unsigned 128-bit immediate integer operand.
+    ///
+    /// This operand is used to pass entire 128-bit vectors as immediates to instructions like
+    /// const.
+    pub uimm128: OperandKind,
 
-    // An unsigned 128-bit immediate integer operand.
-    //
-    // This operand is used to pass entire 128-bit vectors as immediates to
-    // instructions like const.
-    let uimm128 = Builder::new_imm("uimm128")
-        .doc("A 128-bit immediate unsigned integer.")
-        .rust_type("ir::Constant")
-        .build();
-    kinds.push(uimm128);
+    /// A 32-bit immediate signed offset.
+    ///
+    /// This is used to represent an immediate address offset in load/store instructions.
+    pub offset32: OperandKind,
 
-    // A 32-bit immediate signed offset.
-    //
-    // This is used to represent an immediate address offset in load/store
-    // instructions.
-    let offset32 = Builder::new_imm("offset32")
-        .doc("A 32-bit immediate signed offset.")
-        .default_member("offset")
-        .build();
-    kinds.push(offset32);
+    /// A 32-bit immediate floating point operand.
+    ///
+    /// IEEE 754-2008 binary32 interchange format.
+    pub ieee32: OperandKind,
 
-    // A 32-bit immediate floating point operand.
-    //
-    // IEEE 754-2008 binary32 interchange format.
-    let ieee32 = Builder::new_imm("ieee32")
-        .doc("A 32-bit immediate floating point number.")
-        .build();
-    kinds.push(ieee32);
+    /// A 64-bit immediate floating point operand.
+    ///
+    /// IEEE 754-2008 binary64 interchange format.
+    pub ieee64: OperandKind,
 
-    // A 64-bit immediate floating point operand.
-    //
-    // IEEE 754-2008 binary64 interchange format.
-    let ieee64 = Builder::new_imm("ieee64")
-        .doc("A 64-bit immediate floating point number.")
-        .build();
-    kinds.push(ieee64);
+    /// An immediate boolean operand.
+    ///
+    /// This type of immediate boolean can interact with SSA values with any BoolType type.
+    pub boolean: OperandKind,
 
-    // An immediate boolean operand.
-    //
-    // This type of immediate boolean can interact with SSA values with any
-    // BoolType type.
-    let boolean = Builder::new_imm("boolean")
-        .doc("An immediate boolean.")
-        .rust_type("bool")
-        .build();
-    kinds.push(boolean);
+    /// A condition code for comparing integer values.
+    ///
+    /// This enumerated operand kind is used for the `icmp` instruction and corresponds to the
+    /// condcodes::IntCC` Rust type.
+    pub intcc: OperandKind,
 
-    // A condition code for comparing integer values.
-    // This enumerated operand kind is used for the `icmp` instruction and corresponds to the
-    // condcodes::IntCC` Rust type.
-    let mut intcc_values = HashMap::new();
-    intcc_values.insert("eq", "Equal");
-    intcc_values.insert("ne", "NotEqual");
-    intcc_values.insert("sge", "SignedGreaterThanOrEqual");
-    intcc_values.insert("sgt", "SignedGreaterThan");
-    intcc_values.insert("sle", "SignedLessThanOrEqual");
-    intcc_values.insert("slt", "SignedLessThan");
-    intcc_values.insert("uge", "UnsignedGreaterThanOrEqual");
-    intcc_values.insert("ugt", "UnsignedGreaterThan");
-    intcc_values.insert("ule", "UnsignedLessThanOrEqual");
-    intcc_values.insert("ult", "UnsignedLessThan");
-    let intcc = Builder::new_enum("intcc", intcc_values)
-        .doc("An integer comparison condition code.")
-        .default_member("cond")
-        .rust_type("ir::condcodes::IntCC")
-        .build();
-    kinds.push(intcc);
+    /// A condition code for comparing floating point values.
+    ///
+    /// This enumerated operand kind is used for the `fcmp` instruction and corresponds to the
+    /// `condcodes::FloatCC` Rust type.
+    pub floatcc: OperandKind,
 
-    // A condition code for comparing floating point values.  This enumerated operand kind is used
-    // for the `fcmp` instruction and corresponds to the `condcodes::FloatCC` Rust type.
-    let mut floatcc_values = HashMap::new();
-    floatcc_values.insert("ord", "Ordered");
-    floatcc_values.insert("uno", "Unordered");
-    floatcc_values.insert("eq", "Equal");
-    floatcc_values.insert("ne", "NotEqual");
-    floatcc_values.insert("one", "OrderedNotEqual");
-    floatcc_values.insert("ueq", "UnorderedOrEqual");
-    floatcc_values.insert("lt", "LessThan");
-    floatcc_values.insert("le", "LessThanOrEqual");
-    floatcc_values.insert("gt", "GreaterThan");
-    floatcc_values.insert("ge", "GreaterThanOrEqual");
-    floatcc_values.insert("ult", "UnorderedOrLessThan");
-    floatcc_values.insert("ule", "UnorderedOrLessThanOrEqual");
-    floatcc_values.insert("ugt", "UnorderedOrGreaterThan");
-    floatcc_values.insert("uge", "UnorderedOrGreaterThanOrEqual");
-    let floatcc = Builder::new_enum("floatcc", floatcc_values)
-        .doc("A floating point comparison condition code")
-        .default_member("cond")
-        .rust_type("ir::condcodes::FloatCC")
-        .build();
-    kinds.push(floatcc);
+    /// Flags for memory operations like `load` and `store`.
+    pub memflags: OperandKind,
 
-    // Flags for memory operations like :clif:inst:`load` and :clif:inst:`store`.
-    let memflags = Builder::new_imm("memflags")
-        .doc("Memory operation flags")
-        .default_member("flags")
-        .rust_type("ir::MemFlags")
-        .build();
-    kinds.push(memflags);
+    /// A register unit in the current target ISA.
+    pub regunit: OperandKind,
 
-    // A register unit in the current target ISA.
-    let regunit = Builder::new_imm("regunit")
-        .doc("A register unit in the target ISA")
-        .rust_type("isa::RegUnit")
-        .build();
-    kinds.push(regunit);
-
-    // A trap code indicating the reason for trapping.
-    //
-    // The Rust enum type also has a `User(u16)` variant for user-provided trap
-    // codes.
-    let mut trapcode_values = HashMap::new();
-    trapcode_values.insert("stk_ovf", "StackOverflow");
-    trapcode_values.insert("heap_oob", "HeapOutOfBounds");
-    trapcode_values.insert("int_ovf", "IntegerOverflow");
-    trapcode_values.insert("int_divz", "IntegerDivisionByZero");
-    let trapcode = Builder::new_enum("trapcode", trapcode_values)
-        .doc("A trap reason code.")
-        .default_member("code")
-        .rust_type("ir::TrapCode")
-        .build();
-    kinds.push(trapcode);
-
-    return kinds;
+    /// A trap code indicating the reason for trapping.
+    ///
+    /// The Rust enum type also has a `User(u16)` variant for user-provided trap codes.
+    pub trapcode: OperandKind,
+}
+
+impl Immediates {
+    pub fn new() -> Self {
+        Self {
+            imm64: Builder::new_imm("imm64")
+                .doc("A 64-bit immediate integer.")
+                .build(),
+
+            uimm8: Builder::new_imm("uimm8")
+                .doc("An 8-bit immediate unsigned integer.")
+                .build(),
+
+            uimm32: Builder::new_imm("uimm32")
+                .doc("A 32-bit immediate unsigned integer.")
+                .build(),
+
+            uimm128: Builder::new_imm("uimm128")
+                .doc("A 128-bit immediate unsigned integer.")
+                .rust_type("ir::Constant")
+                .build(),
+
+            offset32: Builder::new_imm("offset32")
+                .doc("A 32-bit immediate signed offset.")
+                .default_member("offset")
+                .build(),
+
+            ieee32: Builder::new_imm("ieee32")
+                .doc("A 32-bit immediate floating point number.")
+                .build(),
+
+            ieee64: Builder::new_imm("ieee64")
+                .doc("A 64-bit immediate floating point number.")
+                .build(),
+
+            boolean: Builder::new_imm("boolean")
+                .doc("An immediate boolean.")
+                .rust_type("bool")
+                .build(),
+
+            intcc: {
+                let mut intcc_values = HashMap::new();
+                intcc_values.insert("eq", "Equal");
+                intcc_values.insert("ne", "NotEqual");
+                intcc_values.insert("sge", "SignedGreaterThanOrEqual");
+                intcc_values.insert("sgt", "SignedGreaterThan");
+                intcc_values.insert("sle", "SignedLessThanOrEqual");
+                intcc_values.insert("slt", "SignedLessThan");
+                intcc_values.insert("uge", "UnsignedGreaterThanOrEqual");
+                intcc_values.insert("ugt", "UnsignedGreaterThan");
+                intcc_values.insert("ule", "UnsignedLessThanOrEqual");
+                intcc_values.insert("ult", "UnsignedLessThan");
+                Builder::new_enum("intcc", intcc_values)
+                    .doc("An integer comparison condition code.")
+                    .default_member("cond")
+                    .rust_type("ir::condcodes::IntCC")
+                    .build()
+            },
+
+            floatcc: {
+                let mut floatcc_values = HashMap::new();
+                floatcc_values.insert("ord", "Ordered");
+                floatcc_values.insert("uno", "Unordered");
+                floatcc_values.insert("eq", "Equal");
+                floatcc_values.insert("ne", "NotEqual");
+                floatcc_values.insert("one", "OrderedNotEqual");
+                floatcc_values.insert("ueq", "UnorderedOrEqual");
+                floatcc_values.insert("lt", "LessThan");
+                floatcc_values.insert("le", "LessThanOrEqual");
+                floatcc_values.insert("gt", "GreaterThan");
+                floatcc_values.insert("ge", "GreaterThanOrEqual");
+                floatcc_values.insert("ult", "UnorderedOrLessThan");
+                floatcc_values.insert("ule", "UnorderedOrLessThanOrEqual");
+                floatcc_values.insert("ugt", "UnorderedOrGreaterThan");
+                floatcc_values.insert("uge", "UnorderedOrGreaterThanOrEqual");
+                Builder::new_enum("floatcc", floatcc_values)
+                    .doc("A floating point comparison condition code")
+                    .default_member("cond")
+                    .rust_type("ir::condcodes::FloatCC")
+                    .build()
+            },
+
+            memflags: Builder::new_imm("memflags")
+                .doc("Memory operation flags")
+                .default_member("flags")
+                .rust_type("ir::MemFlags")
+                .build(),
+
+            regunit: Builder::new_imm("regunit")
+                .doc("A register unit in the target ISA")
+                .rust_type("isa::RegUnit")
+                .build(),
+
+            trapcode: {
+                let mut trapcode_values = HashMap::new();
+                trapcode_values.insert("stk_ovf", "StackOverflow");
+                trapcode_values.insert("heap_oob", "HeapOutOfBounds");
+                trapcode_values.insert("int_ovf", "IntegerOverflow");
+                trapcode_values.insert("int_divz", "IntegerDivisionByZero");
+                Builder::new_enum("trapcode", trapcode_values)
+                    .doc("A trap reason code.")
+                    .default_member("code")
+                    .rust_type("ir::TrapCode")
+                    .build()
+            },
+        }
+    }
 }
diff --git a/cranelift/codegen/meta/src/shared/instructions.rs b/cranelift/codegen/meta/src/shared/instructions.rs
index 85c39d71b5..ad94efeb80 100644
--- a/cranelift/codegen/meta/src/shared/instructions.rs
+++ b/cranelift/codegen/meta/src/shared/instructions.rs
@@ -8,12 +8,13 @@ use crate::cdsl::operands::{create_operand as operand, create_operand_doc as ope
 use crate::cdsl::type_inference::Constraint::WiderOrEq;
 use crate::cdsl::types::{LaneType, ValueType};
 use crate::cdsl::typevar::{Interval, TypeSetBuilder, TypeVar};
+use crate::shared::immediates::Immediates;
 use crate::shared::{types, OperandKinds};
 
 pub fn define(
     all_instructions: &mut AllInstructions,
     format_registry: &FormatRegistry,
-    immediates: &OperandKinds,
+    imm: &Immediates,
     entities: &OperandKinds,
 ) -> InstructionGroup {
     let mut ig = InstructionGroupBuilder::new(
@@ -24,20 +25,6 @@ pub fn define(
     );
 
     // Operand kind shorthands.
-    let intcc = immediates.by_name("intcc");
-    let floatcc = immediates.by_name("floatcc");
-    let trapcode = immediates.by_name("trapcode");
-    let uimm8 = immediates.by_name("uimm8");
-    let uimm32 = immediates.by_name("uimm32");
-    let imm64 = immediates.by_name("imm64");
-    let uimm128 = immediates.by_name("uimm128");
-    let offset32 = immediates.by_name("offset32");
-    let memflags = immediates.by_name("memflags");
-    let ieee32 = immediates.by_name("ieee32");
-    let ieee64 = immediates.by_name("ieee64");
-    let boolean = immediates.by_name("boolean");
-    let regunit = immediates.by_name("regunit");
-
     let ebb = entities.by_name("ebb");
     let jump_table = entities.by_name("jump_table");
     let variable_args = entities.by_name("variable_args");
@@ -142,7 +129,7 @@ pub fn define(
 
     let addr = &operand("addr", iAddr);
     let c = &operand_doc("c", Testable, "Controlling value to test");
-    let Cond = &operand("Cond", intcc);
+    let Cond = &operand("Cond", &imm.intcc);
     let x = &operand("x", iB);
     let y = &operand("y", iB);
     let EBB = &operand_doc("EBB", ebb, "Destination extended basic block");
@@ -253,7 +240,7 @@ pub fn define(
         .is_branch(true),
     );
 
-    let Cond = &operand("Cond", floatcc);
+    let Cond = &operand("Cond", &imm.floatcc);
     let f = &operand("f", fflags);
 
     ig.push(
@@ -300,7 +287,7 @@ pub fn define(
     // These are the instructions which br_table legalizes to: they perform address computations,
     // using pointer-sized integers, so their type variables are more constrained.
     let x = &operand_doc("x", iAddr, "index into jump table");
-    let Size = &operand_doc("Size", uimm8, "Size in bytes");
+    let Size = &operand_doc("Size", &imm.uimm8, "Size in bytes");
 
     ig.push(
         Inst::new(
@@ -365,7 +352,7 @@ pub fn define(
         .can_store(true),
     );
 
-    let code = &operand("code", trapcode);
+    let code = &operand("code", &imm.trapcode);
 
     ig.push(
         Inst::new(
@@ -418,7 +405,7 @@ pub fn define(
         .can_trap(true),
     );
 
-    let Cond = &operand("Cond", intcc);
+    let Cond = &operand("Cond", &imm.intcc);
     let f = &operand("f", iflags);
 
     ig.push(
@@ -432,7 +419,7 @@ pub fn define(
         .can_trap(true),
     );
 
-    let Cond = &operand("Cond", floatcc);
+    let Cond = &operand("Cond", &imm.floatcc);
     let f = &operand("f", fflags);
 
     ig.push(
@@ -539,11 +526,11 @@ pub fn define(
     );
 
     let SS = &operand("SS", stack_slot);
-    let Offset = &operand_doc("Offset", offset32, "Byte offset from base address");
+    let Offset = &operand_doc("Offset", &imm.offset32, "Byte offset from base address");
     let x = &operand_doc("x", Mem, "Value to be stored");
     let a = &operand_doc("a", Mem, "Value loaded");
     let p = &operand("p", iAddr);
-    let MemFlags = &operand("MemFlags", memflags);
+    let MemFlags = &operand("MemFlags", &imm.memflags);
     let args = &operand_doc("args", variable_args, "Address arguments");
 
     ig.push(
@@ -876,7 +863,7 @@ pub fn define(
 
     let x = &operand_doc("x", Mem, "Value to be stored");
     let a = &operand_doc("a", Mem, "Value loaded");
-    let Offset = &operand_doc("Offset", offset32, "In-bounds offset into stack slot");
+    let Offset = &operand_doc("Offset", &imm.offset32, "In-bounds offset into stack slot");
 
     ig.push(
         Inst::new(
@@ -962,7 +949,7 @@ pub fn define(
 
     let H = &operand("H", heap);
     let p = &operand("p", HeapOffset);
-    let Size = &operand_doc("Size", uimm32, "Size in bytes");
+    let Size = &operand_doc("Size", &imm.uimm32, "Size in bytes");
 
     ig.push(
         Inst::new(
@@ -990,7 +977,7 @@ pub fn define(
     );
     let T = &operand("T", table);
     let p = &operand("p", TableOffset);
-    let Offset = &operand_doc("Offset", offset32, "Byte offset from element address");
+    let Offset = &operand_doc("Offset", &imm.offset32, "Byte offset from element address");
 
     ig.push(
         Inst::new(
@@ -1013,7 +1000,7 @@ pub fn define(
         .operands_out(vec![addr]),
     );
 
-    let N = &operand("N", imm64);
+    let N = &operand("N", &imm.imm64);
     let a = &operand_doc("a", Int, "A constant integer scalar or vector value");
 
     ig.push(
@@ -1030,7 +1017,7 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let N = &operand("N", ieee32);
+    let N = &operand("N", &imm.ieee32);
     let a = &operand_doc("a", f32_, "A constant f32 scalar value");
 
     ig.push(
@@ -1046,7 +1033,7 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let N = &operand("N", ieee64);
+    let N = &operand("N", &imm.ieee64);
     let a = &operand_doc("a", f64_, "A constant f64 scalar value");
 
     ig.push(
@@ -1062,7 +1049,7 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let N = &operand("N", boolean);
+    let N = &operand("N", &imm.boolean);
     let a = &operand_doc("a", Bool, "A constant boolean scalar or vector value");
 
     ig.push(
@@ -1079,7 +1066,11 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let N = &operand_doc("N", uimm128, "The 16 immediate bytes of a 128-bit vector");
+    let N = &operand_doc(
+        "N",
+        &imm.uimm128,
+        "The 16 immediate bytes of a 128-bit vector",
+    );
     let a = &operand_doc("a", TxN, "A constant vector value");
 
     ig.push(
@@ -1137,7 +1128,7 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let cc = &operand_doc("cc", intcc, "Controlling condition code");
+    let cc = &operand_doc("cc", &imm.intcc, "Controlling condition code");
     let flags = &operand_doc("flags", iflags, "The machine's flag register");
 
     ig.push(
@@ -1217,8 +1208,8 @@ pub fn define(
         .can_load(true),
     );
 
-    let src = &operand("src", regunit);
-    let dst = &operand("dst", regunit);
+    let src = &operand("src", &imm.regunit);
+    let dst = &operand("dst", &imm.regunit);
 
     ig.push(
         Inst::new(
@@ -1302,7 +1293,7 @@ pub fn define(
         .other_side_effects(true),
     );
 
-    let Offset = &operand_doc("Offset", imm64, "Offset from current stack pointer");
+    let Offset = &operand_doc("Offset", &imm.imm64, "Offset from current stack pointer");
 
     ig.push(
         Inst::new(
@@ -1319,7 +1310,7 @@ pub fn define(
         .other_side_effects(true),
     );
 
-    let Offset = &operand_doc("Offset", imm64, "Offset from current stack pointer");
+    let Offset = &operand_doc("Offset", &imm.imm64, "Offset from current stack pointer");
 
     ig.push(
         Inst::new(
@@ -1498,7 +1489,7 @@ pub fn define(
 
     let x = &operand_doc("x", TxN, "SIMD vector to modify");
     let y = &operand_doc("y", &TxN.lane_of(), "New lane value");
-    let Idx = &operand_doc("Idx", uimm8, "Lane index");
+    let Idx = &operand_doc("Idx", &imm.uimm8, "Lane index");
 
     ig.push(
         Inst::new(
@@ -1532,7 +1523,7 @@ pub fn define(
     );
 
     let a = &operand("a", &Int.as_bool());
-    let Cond = &operand("Cond", intcc);
+    let Cond = &operand("Cond", &imm.intcc);
     let x = &operand("x", Int);
     let y = &operand("y", Int);
 
@@ -1566,7 +1557,7 @@ pub fn define(
 
     let a = &operand("a", b1);
     let x = &operand("x", iB);
-    let Y = &operand("Y", imm64);
+    let Y = &operand("Y", &imm.imm64);
 
     ig.push(
         Inst::new(
@@ -1757,7 +1748,7 @@ pub fn define(
 
     let a = &operand("a", iB);
     let x = &operand("x", iB);
-    let Y = &operand("Y", imm64);
+    let Y = &operand("Y", &imm.imm64);
 
     ig.push(
         Inst::new(
@@ -2092,7 +2083,7 @@ pub fn define(
     );
 
     let x = &operand("x", iB);
-    let Y = &operand("Y", imm64);
+    let Y = &operand("Y", &imm.imm64);
     let a = &operand("a", iB);
 
     ig.push(
@@ -2145,7 +2136,7 @@ pub fn define(
 
     let x = &operand_doc("x", Int, "Scalar or vector value to shift");
     let y = &operand_doc("y", iB, "Number of bits to shift");
-    let Y = &operand("Y", imm64);
+    let Y = &operand("Y", &imm.imm64);
     let a = &operand("a", Int);
 
     ig.push(
@@ -2375,7 +2366,7 @@ pub fn define(
             .simd_lanes(Interval::All)
             .build(),
     );
-    let Cond = &operand("Cond", floatcc);
+    let Cond = &operand("Cond", &imm.floatcc);
     let x = &operand("x", Float);
     let y = &operand("y", Float);
     let a = &operand("a", &Float.as_bool());
@@ -2687,7 +2678,7 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let Cond = &operand("Cond", intcc);
+    let Cond = &operand("Cond", &imm.intcc);
     let f = &operand("f", iflags);
     let a = &operand("a", b1);
 
@@ -2705,7 +2696,7 @@ pub fn define(
         .operands_out(vec![a]),
     );
 
-    let Cond = &operand("Cond", floatcc);
+    let Cond = &operand("Cond", &imm.floatcc);
     let f = &operand("f", fflags);
 
     ig.push(
diff --git a/cranelift/codegen/meta/src/shared/legalize.rs b/cranelift/codegen/meta/src/shared/legalize.rs
index 709dab7962..ce2faa7720 100644
--- a/cranelift/codegen/meta/src/shared/legalize.rs
+++ b/cranelift/codegen/meta/src/shared/legalize.rs
@@ -2,12 +2,11 @@ use crate::cdsl::ast::{var, ExprBuilder, Literal};
 use crate::cdsl::instructions::{Instruction, InstructionGroup};
 use crate::cdsl::xform::{TransformGroupBuilder, TransformGroups};
 
-use crate::shared::OperandKinds;
-
+use crate::shared::immediates::Immediates;
 use crate::shared::types::Float::{F32, F64};
 use crate::shared::types::Int::{I16, I32, I64, I8};
 
-pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformGroups {
+pub fn define(insts: &InstructionGroup, imm: &Immediates) -> TransformGroups {
     let mut narrow = TransformGroupBuilder::new(
         "narrow",
         r#"
@@ -139,12 +138,6 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     expand.custom_legalize(insts.by_name("stack_load"), "expand_stack_load");
     expand.custom_legalize(insts.by_name("stack_store"), "expand_stack_store");
 
-    // List of immediates.
-    let imm64 = immediates.by_name("imm64");
-    let ieee32 = immediates.by_name("ieee32");
-    let ieee64 = immediates.by_name("ieee64");
-    let intcc = immediates.by_name("intcc");
-
     // List of variables to reuse in patterns.
     let x = var("x");
     let y = var("y");
@@ -298,7 +291,7 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     }
 
     for &(int_ty, num) in &[(I8, 24), (I16, 16)] {
-        let imm = Literal::constant(imm64, -num);
+        let imm = Literal::constant(&imm.imm64, -num);
 
         widen.legalize(
             def!(a = clz.int_ty(b)),
@@ -322,7 +315,7 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     }
 
     for &(int_ty, num) in &[(I8, 1 << 8), (I16, 1 << 16)] {
-        let num = Literal::constant(imm64, num);
+        let num = Literal::constant(&imm.imm64, num);
         widen.legalize(
             def!(a = ctz.int_ty(b)),
             vec![
@@ -423,7 +416,7 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
         }
 
         for cc in &["eq", "ne", "ugt", "ult", "uge", "ule"] {
-            let w_cc = Literal::enumerator_for(intcc, cc);
+            let w_cc = Literal::enumerator_for(&imm.intcc, cc);
             widen.legalize(
                 def!(a = icmp_imm.int_ty(w_cc, b, c)),
                 vec![def!(x = uextend.I32(b)), def!(a = icmp_imm(w_cc, x, c))],
@@ -439,7 +432,7 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
         }
 
         for cc in &["sgt", "slt", "sge", "sle"] {
-            let w_cc = Literal::enumerator_for(intcc, cc);
+            let w_cc = Literal::enumerator_for(&imm.intcc, cc);
             widen.legalize(
                 def!(a = icmp_imm.int_ty(w_cc, b, c)),
                 vec![def!(x = sextend.I32(b)), def!(a = icmp_imm(w_cc, x, c))],
@@ -534,7 +527,7 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     }
 
     //# Expand bnot using xor.
-    let minus_one = Literal::constant(imm64, -1);
+    let minus_one = Literal::constant(&imm.imm64, -1);
     expand.legalize(
         def!(a = bnot(x)),
         vec![def!(y = iconst(minus_one)), def!(a = bxor(x, y))],
@@ -543,82 +536,82 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     //# Expand bitrev
     //# Adapted from Stack Overflow.
     //# https://stackoverflow.com/questions/746171/most-efficient-algorithm-for-bit-reversal-from-msb-lsb-to-lsb-msb-in-c
-    let imm64_1 = Literal::constant(imm64, 1);
-    let imm64_2 = Literal::constant(imm64, 2);
-    let imm64_4 = Literal::constant(imm64, 4);
+    let imm64_1 = Literal::constant(&imm.imm64, 1);
+    let imm64_2 = Literal::constant(&imm.imm64, 2);
+    let imm64_4 = Literal::constant(&imm.imm64, 4);
 
     widen.legalize(
         def!(a = bitrev.I8(x)),
         vec![
-            def!(a1 = band_imm(x, Literal::constant(imm64, 0xaa))),
+            def!(a1 = band_imm(x, Literal::constant(&imm.imm64, 0xaa))),
             def!(a2 = ushr_imm(a1, imm64_1)),
-            def!(a3 = band_imm(x, Literal::constant(imm64, 0x55))),
+            def!(a3 = band_imm(x, Literal::constant(&imm.imm64, 0x55))),
             def!(a4 = ishl_imm(a3, imm64_1)),
             def!(b = bor(a2, a4)),
-            def!(b1 = band_imm(b, Literal::constant(imm64, 0xcc))),
+            def!(b1 = band_imm(b, Literal::constant(&imm.imm64, 0xcc))),
             def!(b2 = ushr_imm(b1, imm64_2)),
-            def!(b3 = band_imm(b, Literal::constant(imm64, 0x33))),
+            def!(b3 = band_imm(b, Literal::constant(&imm.imm64, 0x33))),
             def!(b4 = ishl_imm(b3, imm64_2)),
             def!(c = bor(b2, b4)),
-            def!(c1 = band_imm(c, Literal::constant(imm64, 0xf0))),
+            def!(c1 = band_imm(c, Literal::constant(&imm.imm64, 0xf0))),
             def!(c2 = ushr_imm(c1, imm64_4)),
-            def!(c3 = band_imm(c, Literal::constant(imm64, 0x0f))),
+            def!(c3 = band_imm(c, Literal::constant(&imm.imm64, 0x0f))),
             def!(c4 = ishl_imm(c3, imm64_4)),
             def!(a = bor(c2, c4)),
         ],
     );
 
-    let imm64_8 = Literal::constant(imm64, 8);
+    let imm64_8 = Literal::constant(&imm.imm64, 8);
 
     widen.legalize(
         def!(a = bitrev.I16(x)),
         vec![
-            def!(a1 = band_imm(x, Literal::constant(imm64, 0xaaaa))),
+            def!(a1 = band_imm(x, Literal::constant(&imm.imm64, 0xaaaa))),
             def!(a2 = ushr_imm(a1, imm64_1)),
-            def!(a3 = band_imm(x, Literal::constant(imm64, 0x5555))),
+            def!(a3 = band_imm(x, Literal::constant(&imm.imm64, 0x5555))),
             def!(a4 = ishl_imm(a3, imm64_1)),
             def!(b = bor(a2, a4)),
-            def!(b1 = band_imm(b, Literal::constant(imm64, 0xcccc))),
+            def!(b1 = band_imm(b, Literal::constant(&imm.imm64, 0xcccc))),
             def!(b2 = ushr_imm(b1, imm64_2)),
-            def!(b3 = band_imm(b, Literal::constant(imm64, 0x3333))),
+            def!(b3 = band_imm(b, Literal::constant(&imm.imm64, 0x3333))),
             def!(b4 = ishl_imm(b3, imm64_2)),
             def!(c = bor(b2, b4)),
-            def!(c1 = band_imm(c, Literal::constant(imm64, 0xf0f0))),
+            def!(c1 = band_imm(c, Literal::constant(&imm.imm64, 0xf0f0))),
             def!(c2 = ushr_imm(c1, imm64_4)),
-            def!(c3 = band_imm(c, Literal::constant(imm64, 0x0f0f))),
+            def!(c3 = band_imm(c, Literal::constant(&imm.imm64, 0x0f0f))),
             def!(c4 = ishl_imm(c3, imm64_4)),
             def!(d = bor(c2, c4)),
-            def!(d1 = band_imm(d, Literal::constant(imm64, 0xff00))),
+            def!(d1 = band_imm(d, Literal::constant(&imm.imm64, 0xff00))),
             def!(d2 = ushr_imm(d1, imm64_8)),
-            def!(d3 = band_imm(d, Literal::constant(imm64, 0x00ff))),
+            def!(d3 = band_imm(d, Literal::constant(&imm.imm64, 0x00ff))),
             def!(d4 = ishl_imm(d3, imm64_8)),
             def!(a = bor(d2, d4)),
         ],
     );
 
-    let imm64_16 = Literal::constant(imm64, 16);
+    let imm64_16 = Literal::constant(&imm.imm64, 16);
 
     expand.legalize(
         def!(a = bitrev.I32(x)),
         vec![
-            def!(a1 = band_imm(x, Literal::constant(imm64, 0xaaaaaaaa))),
+            def!(a1 = band_imm(x, Literal::constant(&imm.imm64, 0xaaaaaaaa))),
             def!(a2 = ushr_imm(a1, imm64_1)),
-            def!(a3 = band_imm(x, Literal::constant(imm64, 0x55555555))),
+            def!(a3 = band_imm(x, Literal::constant(&imm.imm64, 0x55555555))),
             def!(a4 = ishl_imm(a3, imm64_1)),
             def!(b = bor(a2, a4)),
-            def!(b1 = band_imm(b, Literal::constant(imm64, 0xcccccccc))),
+            def!(b1 = band_imm(b, Literal::constant(&imm.imm64, 0xcccccccc))),
             def!(b2 = ushr_imm(b1, imm64_2)),
-            def!(b3 = band_imm(b, Literal::constant(imm64, 0x33333333))),
+            def!(b3 = band_imm(b, Literal::constant(&imm.imm64, 0x33333333))),
             def!(b4 = ishl_imm(b3, imm64_2)),
             def!(c = bor(b2, b4)),
-            def!(c1 = band_imm(c, Literal::constant(imm64, 0xf0f0f0f0))),
+            def!(c1 = band_imm(c, Literal::constant(&imm.imm64, 0xf0f0f0f0))),
             def!(c2 = ushr_imm(c1, imm64_4)),
-            def!(c3 = band_imm(c, Literal::constant(imm64, 0x0f0f0f0f))),
+            def!(c3 = band_imm(c, Literal::constant(&imm.imm64, 0x0f0f0f0f))),
             def!(c4 = ishl_imm(c3, imm64_4)),
             def!(d = bor(c2, c4)),
-            def!(d1 = band_imm(d, Literal::constant(imm64, 0xff00ff00))),
+            def!(d1 = band_imm(d, Literal::constant(&imm.imm64, 0xff00ff00))),
             def!(d2 = ushr_imm(d1, imm64_8)),
-            def!(d3 = band_imm(d, Literal::constant(imm64, 0x00ff00ff))),
+            def!(d3 = band_imm(d, Literal::constant(&imm.imm64, 0x00ff00ff))),
             def!(d4 = ishl_imm(d3, imm64_8)),
             def!(e = bor(d2, d4)),
             def!(e1 = ushr_imm(e, imm64_16)),
@@ -628,21 +621,21 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     );
 
     #[allow(overflowing_literals)]
-    let imm64_0xaaaaaaaaaaaaaaaa = Literal::constant(imm64, 0xaaaaaaaaaaaaaaaa);
-    let imm64_0x5555555555555555 = Literal::constant(imm64, 0x5555555555555555);
+    let imm64_0xaaaaaaaaaaaaaaaa = Literal::constant(&imm.imm64, 0xaaaaaaaaaaaaaaaa);
+    let imm64_0x5555555555555555 = Literal::constant(&imm.imm64, 0x5555555555555555);
     #[allow(overflowing_literals)]
-    let imm64_0xcccccccccccccccc = Literal::constant(imm64, 0xcccccccccccccccc);
-    let imm64_0x3333333333333333 = Literal::constant(imm64, 0x3333333333333333);
+    let imm64_0xcccccccccccccccc = Literal::constant(&imm.imm64, 0xcccccccccccccccc);
+    let imm64_0x3333333333333333 = Literal::constant(&imm.imm64, 0x3333333333333333);
     #[allow(overflowing_literals)]
-    let imm64_0xf0f0f0f0f0f0f0f0 = Literal::constant(imm64, 0xf0f0f0f0f0f0f0f0);
-    let imm64_0x0f0f0f0f0f0f0f0f = Literal::constant(imm64, 0x0f0f0f0f0f0f0f0f);
+    let imm64_0xf0f0f0f0f0f0f0f0 = Literal::constant(&imm.imm64, 0xf0f0f0f0f0f0f0f0);
+    let imm64_0x0f0f0f0f0f0f0f0f = Literal::constant(&imm.imm64, 0x0f0f0f0f0f0f0f0f);
     #[allow(overflowing_literals)]
-    let imm64_0xff00ff00ff00ff00 = Literal::constant(imm64, 0xff00ff00ff00ff00);
-    let imm64_0x00ff00ff00ff00ff = Literal::constant(imm64, 0x00ff00ff00ff00ff);
+    let imm64_0xff00ff00ff00ff00 = Literal::constant(&imm.imm64, 0xff00ff00ff00ff00);
+    let imm64_0x00ff00ff00ff00ff = Literal::constant(&imm.imm64, 0x00ff00ff00ff00ff);
     #[allow(overflowing_literals)]
-    let imm64_0xffff0000ffff0000 = Literal::constant(imm64, 0xffff0000ffff0000);
-    let imm64_0x0000ffff0000ffff = Literal::constant(imm64, 0x0000ffff0000ffff);
-    let imm64_32 = Literal::constant(imm64, 32);
+    let imm64_0xffff0000ffff0000 = Literal::constant(&imm.imm64, 0xffff0000ffff0000);
+    let imm64_0x0000ffff0000ffff = Literal::constant(&imm.imm64, 0x0000ffff0000ffff);
+    let imm64_32 = Literal::constant(&imm.imm64, 32);
 
     expand.legalize(
         def!(a = bitrev.I64(x)),
@@ -680,8 +673,12 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
 
     // Floating-point sign manipulations.
     for &(ty, const_inst, minus_zero) in &[
-        (F32, f32const, &Literal::bits(ieee32, 0x80000000)),
-        (F64, f64const, &Literal::bits(ieee64, 0x8000000000000000)),
+        (F32, f32const, &Literal::bits(&imm.ieee32, 0x80000000)),
+        (
+            F64,
+            f64const,
+            &Literal::bits(&imm.ieee64, 0x8000000000000000),
+        ),
     ] {
         expand.legalize(
             def!(a = fabs.ty(x)),
@@ -724,9 +721,9 @@ pub fn define(insts: &InstructionGroup, immediates: &OperandKinds) -> TransformG
     )
     .chain_with(expand_id);
 
-    let imm64_0 = Literal::constant(imm64, 0);
-    let intcc_ne = Literal::enumerator_for(intcc, "ne");
-    let intcc_eq = Literal::enumerator_for(intcc, "eq");
+    let imm64_0 = Literal::constant(&imm.imm64, 0);
+    let intcc_ne = Literal::enumerator_for(&imm.intcc, "ne");
+    let intcc_eq = Literal::enumerator_for(&imm.intcc, "eq");
 
     expand_flags.legalize(
         def!(trapnz(x, c)),
diff --git a/cranelift/codegen/meta/src/shared/mod.rs b/cranelift/codegen/meta/src/shared/mod.rs
index ba1425b844..837bf3c71d 100644
--- a/cranelift/codegen/meta/src/shared/mod.rs
+++ b/cranelift/codegen/meta/src/shared/mod.rs
@@ -14,11 +14,13 @@ use crate::cdsl::operands::OperandKind;
 use crate::cdsl::settings::SettingGroup;
 use crate::cdsl::xform::TransformGroups;
 
+use crate::shared::immediates::Immediates;
+
 pub struct Definitions {
     pub settings: SettingGroup,
     pub all_instructions: AllInstructions,
     pub instructions: InstructionGroup,
-    pub operand_kinds: OperandKinds,
+    pub imm: Immediates,
     pub format_registry: FormatRegistry,
     pub transform_groups: TransformGroups,
 }
@@ -26,28 +28,12 @@ pub struct Definitions {
 pub struct OperandKinds(Vec<OperandKind>);
 
 impl OperandKinds {
-    pub fn new() -> Self {
-        Self(Vec::new())
-    }
-
     pub fn by_name(&self, name: &'static str) -> &OperandKind {
         self.0
             .iter()
             .find(|op| op.name == name)
             .expect(&format!("unknown Operand name: {}", name))
     }
-
-    pub fn push(&mut self, operand_kind: OperandKind) {
-        assert!(
-            self.0
-                .iter()
-                .find(|existing| existing.name == operand_kind.name)
-                .is_none(),
-            "trying to insert operand kind '{}' for the second time",
-            operand_kind.name
-        );
-        self.0.push(operand_kind);
-    }
 }
 
 impl From<Vec<OperandKind>> for OperandKinds {
@@ -59,7 +45,7 @@ impl From<Vec<OperandKind>> for OperandKinds {
 pub fn define() -> Definitions {
     let mut all_instructions = AllInstructions::new();
 
-    let immediates = OperandKinds(immediates::define());
+    let immediates = Immediates::new();
     let entities = OperandKinds(entities::define());
     let format_registry = formats::define(&immediates, &entities);
     let instructions = instructions::define(
@@ -74,7 +60,7 @@ pub fn define() -> Definitions {
         settings: settings::define(),
         all_instructions,
         instructions,
-        operand_kinds: immediates,
+        imm: immediates,
         format_registry,
         transform_groups,
     }