diff --git a/cranelift/codegen/meta/src/shared/instructions.rs b/cranelift/codegen/meta/src/shared/instructions.rs
index 79797fff4c..cdba177578 100644
--- a/cranelift/codegen/meta/src/shared/instructions.rs
+++ b/cranelift/codegen/meta/src/shared/instructions.rs
@@ -3298,21 +3298,10 @@ pub(crate) fn define(
         .operands_out(vec![a]),
     );
 
-    let Bool = &TypeVar::new(
-        "Bool",
-        "A scalar or vector boolean type",
-        TypeSetBuilder::new()
-            .bools(Interval::All)
-            .simd_lanes(Interval::All)
-            .build(),
-    );
     let BoolTo = &TypeVar::new(
         "BoolTo",
-        "A larger boolean type with the same number of lanes",
-        TypeSetBuilder::new()
-            .bools(Interval::All)
-            .simd_lanes(Interval::All)
-            .build(),
+        "A larger boolean type",
+        TypeSetBuilder::new().bools(Interval::All).build(),
     );
     let x = &Operand::new("x", Bool);
     let a = &Operand::new("a", BoolTo);
@@ -3321,11 +3310,7 @@ pub(crate) fn define(
         Inst::new(
             "bextend",
             r#"
-        Convert `x` to a larger boolean type in the platform-defined way.
-
-        The result type must have the same number of vector lanes as the input,
-        and each lane must not have fewer bits that the input lanes. If the
-        input and output types are the same, this is a no-op.
+        Convert `x` to a larger boolean type
         "#,
             &formats.unary,
         )
@@ -3355,6 +3340,14 @@ pub(crate) fn define(
         .operands_out(vec![a]),
     );
 
+    let Bool = &TypeVar::new(
+        "Bool",
+        "A scalar or vector boolean type",
+        TypeSetBuilder::new()
+            .bools(Interval::All)
+            .simd_lanes(Interval::All)
+            .build(),
+    );
     let IntTo = &TypeVar::new(
         "IntTo",
         "An integer type with the same number of lanes",
diff --git a/cranelift/codegen/src/isa/aarch64/lower.isle b/cranelift/codegen/src/isa/aarch64/lower.isle
index 4730f3262a..feb8eee83e 100644
--- a/cranelift/codegen/src/isa/aarch64/lower.isle
+++ b/cranelift/codegen/src/isa/aarch64/lower.isle
@@ -1124,6 +1124,46 @@
 (rule (lower (bint x))
       (and_imm $I32 x (u64_into_imm_logic $I32 1)))
 
+;;;; Rules for `bmask`/`bextend` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+;; Bextend and Bmask both simply sign-extend. This works for:
+;; - Bextend, because booleans are stored as 0 / -1, so we
+;;   sign-extend the -1 to a -1 in the wider width.
+;; - Bmask, because the resulting integer mask value must be
+;;   all-ones (-1) if the argument is true.
+
+;; Use a common helper to type cast bools to either bool or integer types.
+(decl cast_bool (Type Type Value) InstOutput)
+(rule (lower (has_type out_ty (bextend x @ (value_type in_ty))))
+      (cast_bool in_ty out_ty x))
+(rule (lower (has_type out_ty (bmask x @ (value_type in_ty))))
+      (cast_bool in_ty out_ty x))
+
+
+;; If the target has the same or a smaller size than the source, it's a no-op.
+(rule (cast_bool $B8 $I8 x) x)
+(rule (cast_bool $B16 (fits_in_16 _out) x) x)
+(rule (cast_bool $B32 (fits_in_32 _out) x) x)
+(rule (cast_bool $B64 (fits_in_64 _out) x) x)
+
+;; Casting between 128 bits is a noop
+(rule (cast_bool (ty_int_bool_128 _in) (ty_int_bool_128 _out) x)
+    x)
+
+;; Converting from 128 bits to anything below we just ignore the top register
+(rule (cast_bool (ty_int_bool_128 _in) (fits_in_64 _out) x)
+    (value_regs_get x 0))
+
+;; Extend to 64 bits first, then this will be all 0s or all 1s and we can
+;; duplicate to both halves of 128 bits
+(rule (cast_bool in (ty_int_bool_128 _out) x)
+      (let ((tmp Reg (extend x $true (ty_bits in) 64)))
+        (value_regs tmp tmp)))
+
+;; Values that fit in a single register are sign extended normally
+(rule (cast_bool (fits_in_64 in) (fits_in_64 out) x)
+      (extend x $true (ty_bits in) (ty_bits out)))
+
 ;;;; Rules for `popcnt` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 ;; The implementation of `popcnt` for scalar types is done by moving the value
diff --git a/cranelift/codegen/src/isa/aarch64/lower_inst.rs b/cranelift/codegen/src/isa/aarch64/lower_inst.rs
index 85f1803086..e255af98d6 100644
--- a/cranelift/codegen/src/isa/aarch64/lower_inst.rs
+++ b/cranelift/codegen/src/isa/aarch64/lower_inst.rs
@@ -425,44 +425,7 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
 
         Opcode::Breduce | Opcode::Ireduce => implemented_in_isle(ctx),
 
-        Opcode::Bextend | Opcode::Bmask => {
-            // Bextend and Bmask both simply sign-extend. This works for:
-            // - Bextend, because booleans are stored as 0 / -1, so we
-            //   sign-extend the -1 to a -1 in the wider width.
-            // - Bmask, because the resulting integer mask value must be
-            //   all-ones (-1) if the argument is true.
-
-            let from_ty = ctx.input_ty(insn, 0);
-            let to_ty = ctx.output_ty(insn, 0);
-            let from_bits = ty_bits(from_ty);
-            let to_bits = ty_bits(to_ty);
-
-            if from_ty.is_vector() || from_bits > 64 || to_bits > 64 {
-                return Err(CodegenError::Unsupported(format!(
-                    "{}: Unsupported type: {:?}",
-                    op, from_ty
-                )));
-            }
-
-            assert!(from_bits <= to_bits);
-
-            let rd = get_output_reg(ctx, outputs[0]).only_reg().unwrap();
-            let rn = put_input_in_reg(ctx, inputs[0], NarrowValueMode::None);
-
-            if from_bits == to_bits {
-                ctx.emit(Inst::gen_move(rd, rn, to_ty));
-            } else {
-                let to_bits = if to_bits > 32 { 64 } else { 32 };
-                let from_bits = from_bits as u8;
-                ctx.emit(Inst::Extend {
-                    rd,
-                    rn,
-                    signed: true,
-                    from_bits,
-                    to_bits,
-                });
-            }
-        }
+        Opcode::Bextend | Opcode::Bmask => implemented_in_isle(ctx),
 
         Opcode::Bint => implemented_in_isle(ctx),
 
diff --git a/cranelift/filetests/filetests/isa/aarch64/bitops.clif b/cranelift/filetests/filetests/isa/aarch64/bitops.clif
index da88426aab..c45d89ebd9 100644
--- a/cranelift/filetests/filetests/isa/aarch64/bitops.clif
+++ b/cranelift/filetests/filetests/isa/aarch64/bitops.clif
@@ -317,8 +317,8 @@ block0:
 }
 
 ; block0:
-;   movz x2, #255
-;   sxtb w0, w2
+;   movz x1, #255
+;   sxtb w0, w1
 ;   ret
 
 function %bextend_b1() -> b32 {
@@ -329,8 +329,8 @@ block0:
 }
 
 ; block0:
-;   movz x2, #1
-;   sbfx w0, w2, #0, #1
+;   movz x1, #1
+;   sbfx w0, w1, #0, #1
 ;   ret
 
 function %bnot_i32(i32) -> i32 {
diff --git a/cranelift/filetests/filetests/runtests/bmask.clif b/cranelift/filetests/filetests/runtests/bmask.clif
index d68e59ec00..fb87c02166 100644
--- a/cranelift/filetests/filetests/runtests/bmask.clif
+++ b/cranelift/filetests/filetests/runtests/bmask.clif
@@ -1,4 +1,7 @@
 test interpret
+test run
+target aarch64
+target s390x
 
 function %bmask_b64_i64(b64) -> i64 {
 block0(v0: b64):
diff --git a/cranelift/filetests/filetests/runtests/i128-bextend.clif b/cranelift/filetests/filetests/runtests/i128-bextend.clif
index 34372f98e4..4d53917932 100644
--- a/cranelift/filetests/filetests/runtests/i128-bextend.clif
+++ b/cranelift/filetests/filetests/runtests/i128-bextend.clif
@@ -1,4 +1,6 @@
 test interpret
+test run
+target aarch64
 
 function %bextend_b1_b128(b1) -> b128 {
 block0(v0: b1):
diff --git a/cranelift/filetests/filetests/runtests/i128-bmask.clif b/cranelift/filetests/filetests/runtests/i128-bmask.clif
index f87df7f2ab..2d282ccce9 100644
--- a/cranelift/filetests/filetests/runtests/i128-bmask.clif
+++ b/cranelift/filetests/filetests/runtests/i128-bmask.clif
@@ -1,4 +1,6 @@
 test interpret
+test run
+target aarch64
 
 function %bmask_b128_i128(b128) -> i128 {
 block0(v0: b128):