diff --git a/example/example.rs b/example/example.rs
index 885e55bc764..1ef2aa5dd8e 100644
--- a/example/example.rs
+++ b/example/example.rs
@@ -149,7 +149,7 @@ pub fn array_as_slice(arr: &[u8; 3]) -> &[u8] {
     arr
 }
 
-pub unsafe fn use_ctlz_nonzero(a: u16) -> u16 {
+pub unsafe fn use_ctlz_nonzero(a: u16) -> u32 {
     intrinsics::ctlz_nonzero(a)
 }
 
diff --git a/example/mini_core.rs b/example/mini_core.rs
index e45c16ee280..5e535ff62e1 100644
--- a/example/mini_core.rs
+++ b/example/mini_core.rs
@@ -627,7 +627,7 @@ pub mod intrinsics {
         pub fn min_align_of_val<T: ?::Sized>(val: *const T) -> usize;
         pub fn copy<T>(src: *const T, dst: *mut T, count: usize);
         pub fn transmute<T, U>(e: T) -> U;
-        pub fn ctlz_nonzero<T>(x: T) -> T;
+        pub fn ctlz_nonzero<T>(x: T) -> u32;
         #[rustc_safe_intrinsic]
         pub fn needs_drop<T: ?::Sized>() -> bool;
         #[rustc_safe_intrinsic]
diff --git a/src/intrinsics/mod.rs b/src/intrinsics/mod.rs
index 0b213ff8269..79a90507fa2 100644
--- a/src/intrinsics/mod.rs
+++ b/src/intrinsics/mod.rs
@@ -26,6 +26,7 @@ use rustc_span::source_map::Spanned;
 use rustc_span::symbol::{sym, Symbol};
 
 pub(crate) use self::llvm::codegen_llvm_intrinsic_call;
+use crate::cast::clif_intcast;
 use crate::prelude::*;
 
 fn bug_on_incorrect_arg_count(intrinsic: impl std::fmt::Display) -> ! {
@@ -627,7 +628,8 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             // FIXME trap on `ctlz_nonzero` with zero arg.
             let res = fx.bcx.ins().clz(val);
-            let res = CValue::by_val(res, arg.layout());
+            let res = clif_intcast(fx, res, types::I32, false);
+            let res = CValue::by_val(res, ret.layout());
             ret.write_cvalue(fx, res);
         }
         sym::cttz | sym::cttz_nonzero => {
@@ -636,7 +638,8 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             // FIXME trap on `cttz_nonzero` with zero arg.
             let res = fx.bcx.ins().ctz(val);
-            let res = CValue::by_val(res, arg.layout());
+            let res = clif_intcast(fx, res, types::I32, false);
+            let res = CValue::by_val(res, ret.layout());
             ret.write_cvalue(fx, res);
         }
         sym::ctpop => {
@@ -644,7 +647,8 @@ fn codegen_regular_intrinsic_call<'tcx>(
             let val = arg.load_scalar(fx);
 
             let res = fx.bcx.ins().popcnt(val);
-            let res = CValue::by_val(res, arg.layout());
+            let res = clif_intcast(fx, res, types::I32, false);
+            let res = CValue::by_val(res, ret.layout());
             ret.write_cvalue(fx, res);
         }
         sym::bitreverse => {