diff --git a/docs/langref.rst b/docs/langref.rst
index 5fb42c44a2..cff747752a 100644
--- a/docs/langref.rst
+++ b/docs/langref.rst
@@ -685,7 +685,7 @@ Integer operations
 .. autoinst:: iadd_cout
 .. autoinst:: iadd_carry
 .. autoinst:: isub
-.. autoinst:: isub_imm
+.. autoinst:: irsub_imm
 .. autoinst:: isub_bin
 .. autoinst:: isub_bout
 .. autoinst:: isub_borrow
diff --git a/filetests/parser/tiny.cton b/filetests/parser/tiny.cton
index 588223c521..7158441ed2 100644
--- a/filetests/parser/tiny.cton
+++ b/filetests/parser/tiny.cton
@@ -55,12 +55,14 @@ ebb0(vx0: i32, vx1: i32):
     v0 = icmp eq, vx0, vx1
     v1 = icmp ult, vx0, vx1
     v2 = icmp sge, vx0, vx1
+    v3 = irsub_imm vx1, 45
 }
 ; sameln: function icmp(i32, i32) {
 ; nextln: ebb0(vx0: i32, vx1: i32):
 ; nextln:     v0 = icmp eq, vx0, vx1
 ; nextln:     v1 = icmp ult, vx0, vx1
 ; nextln:     v2 = icmp sge, vx0, vx1
+; nextln:     v3 = irsub_imm vx1, 45
 ; nextln: }
 
 ; Floating condition codes.
diff --git a/lib/cretonne/meta/base/formats.py b/lib/cretonne/meta/base/formats.py
index c72d5034f8..7ed047c330 100644
--- a/lib/cretonne/meta/base/formats.py
+++ b/lib/cretonne/meta/base/formats.py
@@ -22,7 +22,6 @@ UnarySplit = InstructionFormat(VALUE, multiple_results=True)
 
 Binary = InstructionFormat(VALUE, VALUE)
 BinaryImm = InstructionFormat(VALUE, imm64)
-BinaryImmRev = InstructionFormat(imm64, VALUE)
 
 # Generate result + overflow flag.
 BinaryOverflow = InstructionFormat(VALUE, VALUE, multiple_results=True)
diff --git a/lib/cretonne/meta/base/instructions.py b/lib/cretonne/meta/base/instructions.py
index a61f98cefe..41b6ff6c05 100644
--- a/lib/cretonne/meta/base/instructions.py
+++ b/lib/cretonne/meta/base/instructions.py
@@ -507,22 +507,18 @@ srem_imm = Instruction(
         allowed. """,
         ins=(x, Y), outs=a)
 
-# Swap x and y for isub_imm.
-X = Operand('X', imm64)
-y = Operand('y', iB)
-
-isub_imm = Instruction(
-        'isub_imm', """
-        Immediate wrapping subtraction: :math:`a := X - y \pmod{2^B}`.
+irsub_imm = Instruction(
+        'irsub_imm', """
+        Immediate reverse wrapping subtraction: :math:`a := Y - x \pmod{2^B}`.
 
-        Also works as integer negation when :math:`X = 0`. Use :inst:`iadd_imm`
+        Also works as integer negation when :math:`Y = 0`. Use :inst:`iadd_imm`
         with a negative immediate operand for the reverse immediate
         subtraction.
 
         Polymorphic over all scalar integer types, but does not support vector
         types.
         """,
-        ins=(X, y), outs=a)
+        ins=(x, Y), outs=a)
 
 #
 # Integer arithmetic with carry and/or borrow.
diff --git a/lib/cretonne/meta/cdsl/formats.py b/lib/cretonne/meta/cdsl/formats.py
index aa7d384b7c..910afb6052 100644
--- a/lib/cretonne/meta/cdsl/formats.py
+++ b/lib/cretonne/meta/cdsl/formats.py
@@ -25,6 +25,14 @@ class InstructionFormat(object):
     produced. Some instructions, like `call`, may have a variable number of
     results.
 
+    The instruction format stores two separate lists of operands: Immediates
+    and values. Immediate operands (including entity references) are
+    represented as explicit members in the `InstructionData` variants. The
+    value operands are stored differently, depending on how many there are.
+    Beyond a certain point, instruction formats switch to an external value
+    list for storing value arguments. Value lists can hold an arbitrary number
+    of values.
+
     All instruction formats must be predefined in the
     :py:mod:`cretonne.formats` module.
 
@@ -45,8 +53,8 @@ class InstructionFormat(object):
         immediate operands.
     """
 
-    # Map (multiple_results, kind, kind, ...) -> InstructionFormat
-    _registry = dict()  # type: Dict[Tuple[bool, Tuple[OperandKind, ...]], InstructionFormat]  # noqa
+    # Map (multiple_results, imm_kinds, num_value_operands) -> format
+    _registry = dict()  # type: Dict[Tuple[bool, Tuple[OperandKind, ...], int, bool], InstructionFormat]  # noqa
 
     # All existing formats.
     all_formats = list()  # type: List[InstructionFormat]
@@ -62,13 +70,11 @@ class InstructionFormat(object):
         # operands are values or variable argument lists. They are all handled
         # specially.
         self.imm_members = list()  # type: List[str]
-        # Operand kinds for the immediate operands.
-        self.imm_kinds = list()  # type: List[OperandKind]
         # The number of value operands stored in the format, or `None` when
         # `has_value_list` is set.
         self.num_value_operands = 0
-
-        sig_kinds = tuple(self._process_member_names(kinds))
+        # Operand kinds for the immediate operands.
+        self.imm_kinds = tuple(self._process_member_names(kinds))
 
         # The typevar_operand argument must point to a 'value' operand.
         self.typevar_operand = kwargs.get('typevar_operand', None)  # type: int
@@ -81,7 +87,10 @@ class InstructionFormat(object):
             self.typevar_operand = 0
 
         # Compute a signature for the global registry.
-        sig = (self.multiple_results, sig_kinds)
+        sig = (
+                self.multiple_results, self.imm_kinds,
+                self.num_value_operands,
+                self.has_value_list)
         if sig in InstructionFormat._registry:
             raise RuntimeError(
                 "Format '{}' has the same signature as existing format '{}'"
@@ -98,9 +107,9 @@ class InstructionFormat(object):
         pair. The member names correspond to members in the Rust
         `InstructionData` data structure.
 
-        Update the fields `num_value_operands`, `imm_kinds`, and `imm_members`.
+        Update the fields `num_value_operands` and `imm_members`.
 
-        Yields the operand kinds.
+        Yields the immediate operand kinds.
         """
         for arg in kinds:
             if isinstance(arg, OperandKind):
@@ -116,16 +125,14 @@ class InstructionFormat(object):
                 # We require a value list for storage of variable arguments.
                 assert self.has_value_list, "Need a value list"
             else:
-                self.imm_kinds.append(k)
                 self.imm_members.append(member)
-
-            yield k
+                yield k
 
     def __str__(self):
         # type: () -> str
         args = ', '.join('{}: {}'.format(m, k)
                          for m, k in zip(self.imm_members, self.imm_kinds))
-        return '{}({}, values={})'.format(
+        return '{}(imms=({}), vals={})'.format(
                 self.name, args, self.num_value_operands)
 
     def __getattr__(self, attr):
@@ -162,7 +169,13 @@ class InstructionFormat(object):
             multiple_results = outs[0].kind == VARIABLE_ARGS
         else:
             multiple_results = len(outs) > 1
-        sig = (multiple_results, tuple(op.kind for op in ins))
+
+        # Construct a signature.
+        imm_kinds = tuple(op.kind for op in ins if op.is_immediate())
+        num_values = sum(1 for op in ins if op.is_value())
+        has_varargs = (VARIABLE_ARGS in tuple(op.kind for op in ins))
+
+        sig = (multiple_results, imm_kinds, num_values, has_varargs)
         if sig not in InstructionFormat._registry:
             raise RuntimeError(
                     "No instruction format matches ins = ({}){}".format(
diff --git a/lib/cretonne/src/ir/instructions.rs b/lib/cretonne/src/ir/instructions.rs
index 4cabfe6698..09167907be 100644
--- a/lib/cretonne/src/ir/instructions.rs
+++ b/lib/cretonne/src/ir/instructions.rs
@@ -145,13 +145,6 @@ pub enum InstructionData {
         arg: Value,
         imm: Imm64,
     },
-    // Same as `BinaryImm`, but the immediate is the left-hand-side operand.
-    BinaryImmRev {
-        opcode: Opcode,
-        ty: Type,
-        arg: Value,
-        imm: Imm64,
-    },
     BinaryOverflow {
         opcode: Opcode,
         ty: Type,
diff --git a/lib/cretonne/src/write.rs b/lib/cretonne/src/write.rs
index 8f120f447a..57d31d2e5c 100644
--- a/lib/cretonne/src/write.rs
+++ b/lib/cretonne/src/write.rs
@@ -237,7 +237,6 @@ fn write_instruction(w: &mut Write,
         UnarySplit { arg, .. } => writeln!(w, " {}", arg),
         Binary { args, .. } => writeln!(w, " {}, {}", args[0], args[1]),
         BinaryImm { arg, imm, .. } => writeln!(w, " {}, {}", arg, imm),
-        BinaryImmRev { imm, arg, .. } => writeln!(w, " {}, {}", imm, arg),
         BinaryOverflow { args, .. } => writeln!(w, " {}, {}", args[0], args[1]),
         Ternary { args, .. } => writeln!(w, " {}, {}, {}", args[0], args[1], args[2]),
         TernaryOverflow { ref data, .. } => writeln!(w, " {}", data),
diff --git a/lib/reader/src/parser.rs b/lib/reader/src/parser.rs
index 72f43a043d..1e5e194785 100644
--- a/lib/reader/src/parser.rs
+++ b/lib/reader/src/parser.rs
@@ -177,7 +177,6 @@ impl<'a> Context<'a> {
                     InstructionData::Unary { ref mut arg, .. } |
                     InstructionData::UnarySplit { ref mut arg, .. } |
                     InstructionData::BinaryImm { ref mut arg, .. } |
-                    InstructionData::BinaryImmRev { ref mut arg, .. } |
                     InstructionData::ExtractLane { ref mut arg, .. } |
                     InstructionData::BranchTable { ref mut arg, .. } => {
                         self.map.rewrite_value(arg, loc)?;
@@ -1368,17 +1367,6 @@ impl<'a> Parser<'a> {
                     imm: rhs,
                 }
             }
-            InstructionFormat::BinaryImmRev => {
-                let lhs = self.match_imm64("expected immediate integer first operand")?;
-                self.match_token(Token::Comma, "expected ',' between operands")?;
-                let rhs = self.match_value("expected SSA value second operand")?;
-                InstructionData::BinaryImmRev {
-                    opcode: opcode,
-                    ty: VOID,
-                    imm: lhs,
-                    arg: rhs,
-                }
-            }
             InstructionFormat::BinaryOverflow => {
                 let lhs = self.match_value("expected SSA value first operand")?;
                 self.match_token(Token::Comma, "expected ',' between operands")?;