diff --git a/cranelift/docs/langref.rst b/cranelift/docs/langref.rst
index 23cc225d7f..5fb42c44a2 100644
--- a/cranelift/docs/langref.rst
+++ b/cranelift/docs/langref.rst
@@ -668,14 +668,9 @@ allocation pass and beyond.
 Vector operations
 -----------------
 
+.. autoinst:: vsplit
+.. autoinst:: vconcat
 .. autoinst:: vselect
-
-.. inst:: a = vbuild x, y, z, ...
-
-    Vector build.
-
-    Build a vector value from the provided lanes.
-
 .. autoinst:: splat
 .. autoinst:: insertlane
 .. autoinst:: extractlane
diff --git a/cranelift/filetests/isa/riscv/abi.cton b/cranelift/filetests/isa/riscv/abi.cton
index e3bafb83e1..bb27ac20ae 100644
--- a/cranelift/filetests/isa/riscv/abi.cton
+++ b/cranelift/filetests/isa/riscv/abi.cton
@@ -21,6 +21,14 @@ function f(i32) {
     sig3 = signature(f64, f64, f64, f64, f64, f64, f64, i64) -> f64
 ; check: sig3 = signature(f64 [%f10], f64 [%f11], f64 [%f12], f64 [%f13], f64 [%f14], f64 [%f15], f64 [%f16], i32 [0], i32 [4]) -> f64 [%f10]
 
+; Splitting vectors.
+    sig4 = signature(i32x4)
+; check: sig4 = signature(i32 [%x10], i32 [%x11], i32 [%x12], i32 [%x13])
+
+; Splitting vectors, then splitting ints.
+    sig5 = signature(i64x4)
+; check: sig5 = signature(i32 [%x10], i32 [%x11], i32 [%x12], i32 [%x13], i32 [%x14], i32 [%x15], i32 [%x16], i32 [%x17])
+
 ebb0(v0: i32):
     return_reg v0
 }
@@ -32,3 +40,17 @@ ebb0(v0: i64):
     v1 = iadd_imm v0, 1
     return v0
 }
+
+function vector_split_args(i64x4) -> i64 {
+ebb0(v0: i64x4):
+; check: $ebb0($(v0al=$VX): i32, $(v0ah=$VX): i32, $(v0bl=$VX): i32, $(v0bh=$VX): i32, $(v0cl=$VX): i32, $(v0ch=$VX): i32, $(v0dl=$VX): i32, $(v0dh=$VX): i32):
+; check: $(v0a=$V) = iconcat_lohi $v0al, $v0ah
+; check: $(v0b=$V) = iconcat_lohi $v0bl, $v0bh
+; check: $(v0ab=$V) = vconcat $v0a, $v0b
+; check: $(v0c=$V) = iconcat_lohi $v0cl, $v0ch
+; check: $(v0d=$V) = iconcat_lohi $v0dl, $v0dh
+; check: $(v0cd=$V) = vconcat $v0c, $v0d
+; check: $(v0abcd=$V) = vconcat $v0ab, $v0cd
+    v1 = iadd v0, v0
+    return v0
+}
diff --git a/lib/cretonne/meta/base/instructions.py b/lib/cretonne/meta/base/instructions.py
index 9cec7d9d58..09cd2c75c7 100644
--- a/lib/cretonne/meta/base/instructions.py
+++ b/lib/cretonne/meta/base/instructions.py
@@ -264,6 +264,40 @@ fill = Instruction(
 # Vector operations
 #
 
+x = Operand('x', TxN, doc='Vector to split')
+lo = Operand('lo', TxN.half_vector(), doc='Low-numbered lanes of `x`')
+hi = Operand('hi', TxN.half_vector(), doc='High-numbered lanes of `x`')
+
+vsplit = Instruction(
+        'vsplit', r"""
+        Split a vector into two halves.
+
+        Split the vector `x` into two separate values, each containing half of
+        the lanes from ``x``. The result may be two scalars if ``x`` only had
+        two lanes.
+        """,
+        ins=x, outs=(lo, hi))
+
+Any128 = TypeVar(
+        'Any128', 'Any scalar or vector type with as most 128 lanes',
+        ints=True, floats=True, bools=True, scalars=True, simd=(1, 128))
+x = Operand('x', Any128, doc='Low-numbered lanes')
+y = Operand('y', Any128, doc='High-numbered lanes')
+a = Operand('a', Any128.double_vector(), doc='Concatenation of `x` and `y`')
+
+vconcat = Instruction(
+        'vconcat', r"""
+        Vector concatenation.
+
+        Return a vector formed by concatenating ``x`` and ``y``. The resulting
+        vector type has twice as many lanes as each of the inputs. The lanes of
+        ``x`` appear as the low-numbered lanes, and the lanes of ``y`` become
+        the high-numbered lanes of ``a``.
+
+        It is possible to form a vector by concatenating two scalars.
+        """,
+        ins=(x, y), outs=a)
+
 c = Operand('c', TxN.as_bool(), doc='Controlling vector')
 x = Operand('x', TxN, doc='Value to use where `c` is true')
 y = Operand('y', TxN, doc='Value to use where `c` is false')
diff --git a/lib/cretonne/meta/cdsl/typevar.py b/lib/cretonne/meta/cdsl/typevar.py
index 580d367f51..908ebbd8cd 100644
--- a/lib/cretonne/meta/cdsl/typevar.py
+++ b/lib/cretonne/meta/cdsl/typevar.py
@@ -315,6 +315,8 @@ class TypeVar(object):
     ASBOOL = 'as_bool'
     HALFWIDTH = 'half_width'
     DOUBLEWIDTH = 'double_width'
+    HALFVECTOR = 'half_vector'
+    DOUBLEVECTOR = 'double_vector'
 
     @staticmethod
     def derived(base, derived_func):
@@ -396,6 +398,30 @@ class TypeVar(object):
 
         return TypeVar.derived(self, self.DOUBLEWIDTH)
 
+    def half_vector(self):
+        # type: () -> TypeVar
+        """
+        Return a derived type variable that has half the number of vector lanes
+        as this one, with the same lane type.
+        """
+        if not self.is_derived:
+            ts = self.type_set
+            assert ts.min_lanes > 1, "Can't halve a scalar type"
+
+        return TypeVar.derived(self, self.HALFVECTOR)
+
+    def double_vector(self):
+        # type: () -> TypeVar
+        """
+        Return a derived type variable that has twice the number of vector
+        lanes as this one, with the same lane type.
+        """
+        if not self.is_derived:
+            ts = self.type_set
+            assert ts.max_lanes < 256, "Can't double 256 lanes."
+
+        return TypeVar.derived(self, self.DOUBLEVECTOR)
+
     def free_typevar(self):
         # type: () -> TypeVar
         """
diff --git a/lib/cretonne/src/ir/instructions.rs b/lib/cretonne/src/ir/instructions.rs
index c4ac3b0c20..2a956ffc86 100644
--- a/lib/cretonne/src/ir/instructions.rs
+++ b/lib/cretonne/src/ir/instructions.rs
@@ -708,6 +708,12 @@ enum OperandConstraint {
 
     /// This operand is `ctrlType.double_width()`.
     DoubleWidth,
+
+    /// This operand is `ctrlType.half_vector()`.
+    HalfVector,
+
+    /// This operand is `ctrlType.double_vector()`.
+    DoubleVector,
 }
 
 impl OperandConstraint {
@@ -725,6 +731,8 @@ impl OperandConstraint {
             AsBool => Some(ctrl_type.as_bool()),
             HalfWidth => Some(ctrl_type.half_width().expect("invalid type for half_width")),
             DoubleWidth => Some(ctrl_type.double_width().expect("invalid type for double_width")),
+            HalfVector => Some(ctrl_type.half_vector().expect("invalid type for half_vector")),
+            DoubleVector => Some(ctrl_type.by(2).expect("invalid type for double_vector")),
         }
     }
 }
diff --git a/lib/cretonne/src/ir/types.rs b/lib/cretonne/src/ir/types.rs
index 2391ca6072..550f25804b 100644
--- a/lib/cretonne/src/ir/types.rs
+++ b/lib/cretonne/src/ir/types.rs
@@ -217,6 +217,8 @@ impl Type {
     }
 
     /// Get a SIMD vector with half the number of lanes.
+    ///
+    /// There is no `double_vector()` method. Use `t.by(2)` instead.
     pub fn half_vector(self) -> Option<Type> {
         if self.is_scalar() {
             None
diff --git a/lib/cretonne/src/legalizer.rs b/lib/cretonne/src/legalizer.rs
index 885a034084..9223a174b0 100644
--- a/lib/cretonne/src/legalizer.rs
+++ b/lib/cretonne/src/legalizer.rs
@@ -173,9 +173,9 @@ fn convert_from_abi(dfg: &mut DataFlowGraph,
         // Construct a `ty` by concatenating two halves of a vector.
         ValueConversion::VectorSplit => {
             let abi_ty = ty.half_vector().expect("Invalid type for conversion");
-            let _lo = convert_from_abi(dfg, pos, entry, abi_arg, abi_types, abi_ty);
-            let _hi = convert_from_abi(dfg, pos, entry, abi_arg, abi_types, abi_ty);
-            unimplemented!()
+            let lo = convert_from_abi(dfg, pos, entry, abi_arg, abi_types, abi_ty);
+            let hi = convert_from_abi(dfg, pos, entry, abi_arg, abi_types, abi_ty);
+            dfg.ins(pos).vconcat(lo, hi)
         }
         // Construct a `ty` by bit-casting from an integer type.
         ValueConversion::IntBits => {