diff --git a/etc/function-definitions.json b/etc/function-definitions.json
index 3cf7e0fed..39b6c9702 100644
--- a/etc/function-definitions.json
+++ b/etc/function-definitions.json
@@ -604,12 +604,16 @@
     "rint": {
         "sources": [
             "src/libm_helper.rs",
+            "src/math/arch/aarch64.rs",
+            "src/math/arch/wasm32.rs",
             "src/math/rint.rs"
         ],
         "type": "f64"
     },
     "rintf": {
         "sources": [
+            "src/math/arch/aarch64.rs",
+            "src/math/arch/wasm32.rs",
             "src/math/rintf.rs"
         ],
         "type": "f32"
diff --git a/src/math/arch/aarch64.rs b/src/math/arch/aarch64.rs
new file mode 100644
index 000000000..374ec11bf
--- /dev/null
+++ b/src/math/arch/aarch64.rs
@@ -0,0 +1,33 @@
+use core::arch::aarch64::{
+    float32x2_t, float64x1_t, vdup_n_f32, vdup_n_f64, vget_lane_f32, vget_lane_f64, vrndn_f32,
+    vrndn_f64,
+};
+
+pub fn rint(x: f64) -> f64 {
+    // SAFETY: only requires target_feature=neon, ensured by `cfg_if` in parent module.
+    let x_vec: float64x1_t = unsafe { vdup_n_f64(x) };
+
+    // SAFETY: only requires target_feature=neon, ensured by `cfg_if` in parent module.
+    let result_vec: float64x1_t = unsafe { vrndn_f64(x_vec) };
+
+    // SAFETY: only requires target_feature=neon, ensured by `cfg_if` in parent module.
+    let result: f64 = unsafe { vget_lane_f64::<0>(result_vec) };
+
+    result
+}
+
+pub fn rintf(x: f32) -> f32 {
+    // There's a scalar form of this instruction (FRINTN) but core::arch doesn't expose it, so we
+    // have to use the vector form and drop the other lanes afterwards.
+
+    // SAFETY: only requires target_feature=neon, ensured by `cfg_if` in parent module.
+    let x_vec: float32x2_t = unsafe { vdup_n_f32(x) };
+
+    // SAFETY: only requires target_feature=neon, ensured by `cfg_if` in parent module.
+    let result_vec: float32x2_t = unsafe { vrndn_f32(x_vec) };
+
+    // SAFETY: only requires target_feature=neon, ensured by `cfg_if` in parent module.
+    let result: f32 = unsafe { vget_lane_f32::<0>(result_vec) };
+
+    result
+}
diff --git a/src/math/arch/mod.rs b/src/math/arch/mod.rs
index bd79ae1c6..091d7650a 100644
--- a/src/math/arch/mod.rs
+++ b/src/math/arch/mod.rs
@@ -11,10 +11,19 @@
 cfg_if! {
     if #[cfg(all(target_arch = "wasm32", intrinsics_enabled))] {
         mod wasm32;
-        pub use wasm32::{ceil, ceilf, fabs, fabsf, floor, floorf, sqrt, sqrtf, trunc, truncf};
+        pub use wasm32::{
+            ceil, ceilf, fabs, fabsf, floor, floorf, rint, rintf, sqrt, sqrtf, trunc, truncf,
+        };
     } else if #[cfg(target_feature = "sse2")] {
         mod i686;
         pub use i686::{sqrt, sqrtf};
+    } else if #[cfg(all(
+        target_arch = "aarch64", // TODO: also arm64ec?
+        target_feature = "neon",
+        target_endian = "little", // see https://github.com/rust-lang/stdarch/issues/1484
+    ))] {
+        mod aarch64;
+        pub use aarch64::{rint, rintf};
     }
 }
 
diff --git a/src/math/arch/wasm32.rs b/src/math/arch/wasm32.rs
index 384445f12..de80c8a58 100644
--- a/src/math/arch/wasm32.rs
+++ b/src/math/arch/wasm32.rs
@@ -25,6 +25,14 @@ pub fn floorf(x: f32) -> f32 {
     core::arch::wasm32::f32_floor(x)
 }
 
+pub fn rint(x: f64) -> f64 {
+    core::arch::wasm32::f64_nearest(x)
+}
+
+pub fn rintf(x: f32) -> f32 {
+    core::arch::wasm32::f32_nearest(x)
+}
+
 pub fn sqrt(x: f64) -> f64 {
     core::arch::wasm32::f64_sqrt(x)
 }
diff --git a/src/math/rint.rs b/src/math/rint.rs
index cbdc3c2b9..c9ea6402e 100644
--- a/src/math/rint.rs
+++ b/src/math/rint.rs
@@ -1,5 +1,14 @@
 #[cfg_attr(all(test, assert_no_panic), no_panic::no_panic)]
 pub fn rint(x: f64) -> f64 {
+    select_implementation! {
+        name: rint,
+        use_arch: any(
+            all(target_arch = "wasm32", intrinsics_enabled),
+            all(target_arch = "aarch64", target_feature = "neon", target_endian = "little"),
+        ),
+        args: x,
+    }
+
     let one_over_e = 1.0 / f64::EPSILON;
     let as_u64: u64 = x.to_bits();
     let exponent: u64 = (as_u64 >> 52) & 0x7ff;
diff --git a/src/math/rintf.rs b/src/math/rintf.rs
index 2d22c9393..33b5b3dde 100644
--- a/src/math/rintf.rs
+++ b/src/math/rintf.rs
@@ -1,5 +1,14 @@
 #[cfg_attr(all(test, assert_no_panic), no_panic::no_panic)]
 pub fn rintf(x: f32) -> f32 {
+    select_implementation! {
+        name: rintf,
+        use_arch: any(
+            all(target_arch = "wasm32", intrinsics_enabled),
+            all(target_arch = "aarch64", target_feature = "neon", target_endian = "little"),
+        ),
+        args: x,
+    }
+
     let one_over_e = 1.0 / f32::EPSILON;
     let as_u32: u32 = x.to_bits();
     let exponent: u32 = (as_u32 >> 23) & 0xff;